# Pack inventory

Which seed packs we need to make real things with the whole system, which exist in
this library, which were added now, and what is still missing. Dated October 29, 2026.

A seed pack is a collection of nouns (reusable things) and verbs (claimable work). Every
pack below is a **proposal**: validated offline against Tessera's models, but not
installed, imported, admitted or run by any worker. "Exists" means it was in this
library before today's owner direction; "Added now" means this change built it;
"Needed" means nobody has built it yet.

Noun and verb kinds are Tessera's registered categories, without the `noun.` and `verb.`
prefixes. "Proposes" lists follow-up verbs a worker returns in `proposed_verbs`; an
authorized requester decides whether to enqueue them.

## 1. Making a video from start to finish

| Step | Pack and verb | Status | Noun kinds | Verb kinds | Proposes |
| --- | --- | --- | --- | --- | --- |
| Script and story | [Video showcase](../packs/video-showcase/README.md): draft-script | Exists | scene_recipe (script), eval_run (evidence links), seed_hypothesis (brief) | solve_draft | storyboard |
| Storyboard and shot list | Video showcase: storyboard | Exists | scene_recipe (storyboard, shot list, capture plan) | solve_draft | capture-screen per capture, draft-scene |
| Screen capture of real products | Video showcase: capture-screen | Exists | scene_recipe (media in the video and motion format), code_node (Playwright, Chromium cards from video and motion) | compile_dag | |
| Narration, a person's verb | Video showcase: record-narration | Exists | scene_recipe (media) | solve_draft (human) | transcribe-captions |
| Transcribed captions | Video showcase: transcribe-captions | Exists | scene_recipe (caption track) | ingest | |
| Motion graphics as code | [Visual and 3D rendering](../packs/visual-rendering/README.md): plan-scene, render; Video and motion: seek-frame, fit-motion, draft-scene | Added now | motion_profile (formulas, traces, profiles), scene_recipe (scene, asset) | solve_draft, compile_dag, ingest | prepare-assets, compile-social-formats |
| Audio layers, sound effects and music with license checks | Video showcase: prepare-audio, asset policy | Added now | skill_pack (asset policy), scene_recipe (audio cue sheet) | compile_dag | |
| Video layers and styles | Video showcase: style presets; [Video and motion](../packs/video-motion/README.md): plan-sequence | Added now | skill_pack (style presets, caption and audio rules), scene_recipe (sequence) | solve_draft | render-sequence |
| Render and encode | Video and motion: draft-scene, compile-social-formats, render-sequence; Visual and 3D rendering: render | Exists, extended; the video and motion tools are tested | eval_run (render manifest), code_node (video and motion tools, FFmpeg, Playwright, Chromium, Blender, three.js) | compile_dag | check-video |
| Every-frame and loudness QA | Video and motion: check-social-formats, check-video; Visual and 3D rendering: check-render; Video showcase: showcase checks in review-video | Exists, extended | benchmark_suite (thresholds, quality checks), eval_run (QA report) | verify_gate | fixes for failed checks |
| Independent review | Video showcase: review-video | Exists | eval_run (review notes) | verify_gate (person or different agent) | |
| Release, a person's verb | Video showcase: record-publication | Added now | eval_run (publication record) | verify_gate (human) | |
| One frame clock for picture, voice and captions | [Narrated explainer video](../packs/narrated-explainer-video/README.md): plan-cut | Added now | scene_recipe (cut plan), reflex_procedure (frame clock rules), code_node (explainer checks) | solve_draft | build-stage-scenes |
| Narration in a synthetic voice, labeled as AI | [Voice narration](../packs/voice-narration/README.md): prepare-lexicon, synthesize-lines, measure-lines, select-takes, place-on-beats, listen-and-approve (a person's verb) | Added now | skill_pack (lexicon), model_spec (voice model card), eval_run (synthesis log, measurements, take choice, listening record), scene_recipe (narration track) | generate_seed, compile_dag, verify_gate | listen-and-approve |
| Scenes as code on an HTML stage, rendered losslessly | Narrated explainer video: build-stage-scenes, render-slices | Added now | scene_recipe (stage), eval_run (slice recordings) | solve_draft, compile_dag | render-slices |
| Captions timed to the voice, as WebVTT, SRT and open captions | Narrated explainer video: time-captions, check-captions (a different worker) | Added now | scene_recipe (caption cues, caption track), eval_run (caption check) | compile_dag, verify_gate | compose-cut |
| Two variants, a frame sheet, every receipt and a hashed release manifest | Narrated explainer video: compose-cut, make-frame-sheet, compile-receipts, assemble-release-manifest | Added now | scene_recipe (sequences), eval_run (compose record, frame sheet, receipts ledger, release manifest) | compile_dag, verify_gate | review-video, record-publication |
| Tool and benchmark freshness for video and motion tools | [Tool discovery](../packs/tool-discovery/README.md), [Benchmark scores](../packs/benchmark-scores/README.md) | Exists, added now | code_node (candidates), eval_run (scores), benchmark_suite (trials) | probe_source, fetch_feed, recheck_fragility, benchmark_harness | run-trial, compare-on-profile |

**Still needed for video:** caption translation into other languages; an audio
description track for blind viewers; and sourcing of licensed music and stock clips.
Narration noise-floor and clipping checks are now in the voice narration pack. Editing,
review, extended QA, cut-downs with platform profiles, covers and titles, and audience
acceptance are in section 8.

## 2. Building agent kits for nonprofits

| Step | Pack and verb | Status | Noun kinds | Verb kinds | Proposes |
| --- | --- | --- | --- | --- | --- |
| The need or brief | [Public-good kits](../packs/public-good-kits/README.md): frame-need | Added now | seed_hypothesis (need brief) | problemize | draft-kit |
| Skill, MCP tool and prompt files | Public-good kits: draft-kit | Added now | skill_pack (manifest, skill file), mcp_server (read-only tool), prompt_shim (prompt file) | solve_draft | write-test-cases, self-check |
| Synthetic test cases | Public-good kits: write-test-cases | Added now | benchmark_suite (test cases) | generate_seed | |
| Self-check | Public-good kits: self-check | Added now | benchmark_suite (self-checks), eval_run (results) | verify_gate | independent-review |
| Independent review by a different worker | Public-good kits: independent-review | Added now | eval_run (review) | verify_gate | |
| Accessibility and translation | Public-good kits: check-accessibility, translate-kit | Added now | benchmark_suite (criteria), skill_pack (translation) | verify_gate, solve_draft | translate-kit per missing language, a fluent review |
| Steward release, a person's verb | Public-good kits: steward-release | Added now | eval_run (release record) | verify_gate (human) | |
| Checking the kit's own code and tools | [Security review](../packs/security-review/README.md), [Checks of checks](../packs/checks-of-checks/README.md) | Exists | seed_hypothesis (findings), benchmark_suite (checks) | verify_gate, solve_draft | |

**Still needed for kits:** feedback and incident intake from organizations using a kit;
a privacy check for kits that touch personal data; upgrade notes when a kit changes
under people already using it; and freshness checks for legal and referral sources,
which need human-sensitive handling and a person in every decision.

## 3. Keeping knowledge fresh

| Step | Pack and verb | Status | Noun kinds | Verb kinds | Proposes |
| --- | --- | --- | --- | --- | --- |
| New-tool discovery | Tool discovery: discover-sources, scan-source, propose-candidate | Exists | public_source, feed_source (sources), code_node (candidates) | probe_source, fetch_feed, generate_seed | run-trial |
| New-repository discovery | Tool discovery: discover-repos | Added now | code_node (candidates) | probe_source | propose-candidate |
| Refreshing tool cards and decisions | Tool discovery: refresh-candidates, daily schedule proposal | Added now | code_node, seed_hypothesis (decision, schedule) | recheck_fragility | run-trial, decide |
| Daily benchmark refresh | Benchmark scores: refresh-scores, flag-stale, daily schedule proposal | Added now | eval_run (scores), reflex_procedure (staleness policy) | recheck_fragility, verify_gate | refresh-scores, compare-on-profile |
| Deprecation and staleness checks | [Freshness checks](../packs/freshness-checks/README.md): scan-staleness, check-deprecations, propose-replacements, record-tombstone | Added now | seed_hypothesis (watch item), public_source (notice), negative_tombstone | recheck_fragility, probe_source, generate_seed, verify_gate | check-deprecations, run-trial, record-tombstone |
| Fair comparisons when something changes | [Evaluation design](../packs/evaluation-design/README.md) | Exists | benchmark_suite, eval_run, seed_hypothesis (pass marks) | problemize, ingest, benchmark_harness, verify_gate | run-arm twice per arm |

**Still needed for freshness:** scoring which sources are worth reading; newsletter
intake; and a person-led verb for resolving contradictions between sources.

## 4. Steering, status, releases and renames

Added October 30, 2026, for work that comes back in almost every project: turning a person's
direction into rules and verbs, answering status questions with evidence, releasing changes and
proving what is live, and renaming things without leaving old names or broken links behind.

| Step | Pack and verb | Status | Noun kinds | Verb kinds | Proposes |
| --- | --- | --- | --- | --- | --- |
| A message becomes typed parts | [Direction to work](../packs/direction-to-work/README.md): split-message | Added now | seed_hypothesis (intake) | situate | draft-directions, record-finding, plan-work, decide-or-ask |
| Directions with scope, source and supersession | Direction to work: draft-directions, find-conflicts, confirm-directions, a person's verb | Added now | seed_hypothesis (direction, conflict, confirmation), reflex_procedure (supersession, term rules, ask policy) | distill_reflex, recheck_fragility, reconcile_conflict (human) | find-conflicts, confirm-directions |
| Requests become atomic verbs; choices are decided or asked once | Direction to work: plan-work, decide-or-ask | Added now | seed_hypothesis (work plan, decision) | decompose_dag, solve_draft | propose-breakdown, record-finding |
| Later work follows the directions | Direction to work: sweep-compliance | Added now | eval_run (compliance report) | verify_gate | record-finding per violation |
| What is done, what is next and where to resume | [Status and handoff](../packs/status-and-handoff/README.md): take-snapshot, write-status, recheck-claims, rank-next-steps, write-handoff, resume-from-handoff | Added now | eval_run (snapshot, status, claim check, resume check), seed_hypothesis (next steps, handoff), recovery_playbook (resume) | probe_source, situate, verify_gate, decompose_dag, solve_draft, recover_fault | record-finding for contradicted claims |
| Releasing a change and proving what is live | [Release and verify](../packs/release-and-verify/README.md): plan-release, check-head, record-release (the release owner's verb), verify-parity, run-smoke, write-receipt, prepare-rollback | Added now | seed_hypothesis (plan, rollback brief), eval_run (head check, record, parity, smoke, receipt), benchmark_suite (smoke checklist) | decompose_dag, verify_gate, benchmark_harness, recover_fault | prepare-rollback, record-finding |
| Renames that reach every reference | [Rename and relink](../packs/rename-and-relink/README.md): build-inventory, plan-rewrites, rewrite-references, record-host-renames (the owner's verb), check-links, sweep-old-names, write-archive-notice | Added now | seed_hypothesis (map, plan), eval_run (inventory, link check, record, sweep), code_node (rewrite), public_source (notice) | probe_source, decompose_dag, solve_draft, verify_gate | record-finding for names that come back |
| A screenshot sent with little or no text, described before anything is judged | [Screenshot design review](../packs/screenshot-design-review/README.md): describe-screenshot | Added now | seed_hypothesis (description) | situate | plan-captures |
| Captures at named widths and themes, and seven design rules measured in every cell | Screenshot design review: plan-captures, capture-screens, check-design-rules | Added now | scene_recipe (capture plan), eval_run (capture set, layout facts, check report), benchmark_suite (design rules) | decompose_dag, compile_dag, verify_gate | record-design-findings |
| Located findings with a proposed fix verb, and the same cells re-rendered and verified by a different worker | Screenshot design review: record-design-findings, rerender-and-verify; [Findings to fixes](../packs/findings-to-fixes/README.md): triage, propose-fix, fix, close | Added now | seed_hypothesis (findings in the findings to fixes shape), eval_run (re-render checks, verifications) | problemize, verify_gate | propose-fix for each finding, a palette brief for a matter of taste |

**Still needed here:** mining past session prompts into categories and reusable verbs under strict
privacy rules; and finishing the most valuable work first when a usage allowance is about to run out.
Review of a screenshot sent with little or no text, with a capture matrix of widths and themes, is
now the screenshot design review pack.

## 5. Research, company research and model tuning

Added October 30, 2026, for topics people asked for by name: research that ends in a cited report,
research on public companies that is never financial advice, and tuning a small model that keeps a
change only when a rule fixed in advance says so. Later that day the options chain review and checked
data analysis packs were added here, and the screenshot design review to section 4.

| Step | Pack and verb | Status | Noun kinds | Verb kinds | Proposes |
| --- | --- | --- | --- | --- | --- |
| A question framed, a search planned for and against the answer | [Deep research report](../packs/deep-research-report/README.md): frame-question, plan-search | Added now | seed_hypothesis (question, search plan), reflex_procedure (access, screening and strength rules) | problemize, expand_seed | collect-sources |
| Sources collected, screened and quoted exactly | Deep research report: collect-sources, screen-sources, extract-claims | Added now | public_source (source log), eval_run (screening log), seed_hypothesis (claims) | probe_source, verify_gate, ingest | synthesize |
| A labeled synthesis, an adversarial review and a report with limits | Deep research report: synthesize, adversarial-review (a different worker), write-report, decide-delivery (a person's verb) | Added now | seed_hypothesis (synthesis, report), eval_run (review), benchmark_suite (report checks) | solve_draft, verify_gate | counter searches the review missed |
| Filings fetched under the SEC's fair-access rules | [Public company research](../packs/public-company-research/README.md): fetch-filings | Added now | public_source (filing index), reflex_procedure (EDGAR access rules) | probe_source | extract-metrics |
| Metrics, comparisons, risk factors and a dated calendar | Public company research: extract-metrics, compare-periods-and-peers, summarize-risk-factors, build-catalyst-calendar | Added now | seed_hypothesis (metric table, comparison, risk factors, calendar) | ingest, situate, solve_draft | fetch-filings for a peer |
| A cited brief, never advice, reviewed and decided by people | Public company research: write-cited-brief, review-brief (a different worker), decide-delivery (a person's verb) | Added now | seed_hypothesis (brief), eval_run (review), benchmark_suite (brief checks) | solve_draft, verify_gate | |
| Owned examples, curated and frozen before any scoring | [Continuous tuning](../packs/continuous-tuning/README.md): register-rule (a person's verb), generate-examples, curate-dataset, freeze-test-set | Added now | code_node (promotion rule), benchmark_suite (batches, frozen test receipts) | verify_gate, solve_draft, ingest | train-adapter |
| An adapter trained, scored, reloaded and decided by the rule | Continuous tuning: train-adapter, evaluate, check-reload, decide-promotion (a different worker), promote-or-roll-back | Added now | model_spec (adapter card, champion pointer), eval_run (runs, reload check, decision, roll back) | solve_draft, benchmark_harness, verify_gate | the next round's generate-examples |
| One timestamped options chain snapshot, read under its source's terms | [Options chain review](../packs/options-chain-review/README.md): capture-chain | Added now | feed_source (snapshot, implied volatility history), reflex_procedure (data access rules) | ingest | compute-measures |
| Greeks beside the provider's, implied volatility and its rank, the person's screens and roll candidates | Options chain review: compute-measures, screen-liquidity, list-roll-candidates | Added now | eval_run (measures, screen), seed_hypothesis (candidates), code_node (chain measures) | verify_gate, situate | write-chain-review |
| A descriptive chain review, never advice, rechecked and decided by people | Options chain review: write-chain-review, check-chain-review (a different worker), decide-delivery (a person's verb) | Added now | seed_hypothesis (review), eval_run (check, delivery record) | solve_draft, verify_gate | |
| A CSV file pinned, profiled, validated and flagged, with no row dropped | [Checked data analysis](../packs/checked-data-analysis/README.md): profile-schema, validate-types-and-ranges, flag-missing-and-outliers | Added now | eval_run (profile, validation, flags), code_node (CSV checks) | ingest, verify_gate | compute-summaries |
| Recorded queries, charts as data and findings traced to queries, rechecked and read by a person | Checked data analysis: compute-summaries, make-chart-specs, write-findings, check-findings (a different worker), decide-meaning (a person's verb) | Added now | eval_run (query results, check, interpretation), scene_recipe (chart specs), seed_hypothesis (findings) | situate, compile_dag, solve_draft, verify_gate | |

**Still needed here:** research intake from newsletters and feeds. The descriptive options chain
review and checked analysis of a CSV file are now packs, added later on October 30.

## 6. Steps, unattended runs, reuse, file triage and claims

Added October 30, 2026, from patterns that recur across design notes, agent sessions and project
folders: what one atomic step should receive, running work with nobody watching, looking up and
reusing before generating, organizing very large collections of files, and notes whose claims say
how they are known.

| Step | Pack and verb | Status | Noun kinds | Verb kinds | Proposes |
| --- | --- | --- | --- | --- | --- |
| A step kind and its grant | [Atomic step contract](../packs/atomic-step-contract/README.md): define-step-kind, audit-grant | Added now | seed_hypothesis (step kind), eval_run (grant audit), reflex_procedure (packet, grant, refusal and patch rules) | distill_reflex, verify_gate | assemble-step-packet |
| One step's packet, its patch and its refusals | Atomic step contract: assemble-step-packet, check-patch, review-refusals, review-pull-log | Added now | seed_hypothesis (packet, revision), eval_run (patch check, refusal review), benchmark_suite (step checks) | situate, verify_gate, recheck_fragility, distill_reflex | record-finding per refusal that names no legal set |
| What a step costs before any content | Atomic step contract: measure-step-overhead | Added now | eval_run (overhead measurement), reflex_procedure (overhead model) | benchmark_harness | define-step-kind for a cheaper configuration |
| A plan approved before a window opens | [Unattended run record](../packs/unattended-run-record/README.md): plan-run, approve-run (a person's verb) | Added now | seed_hypothesis (run plan), eval_run (approval), reflex_procedure (run guards) | decompose_dag, verify_gate (human) | triage-candidates |
| Triage, baselines and attempts in isolated copies | Unattended run record: triage-candidates, take-baseline, attempt-candidate | Added now | seed_hypothesis (candidate), eval_run (triage, baseline, attempt), reflex_procedure (fallback ladder) | problemize, benchmark_harness, solve_draft | grade-attempt |
| Grading by the check, and a morning report | Unattended run record: grade-attempt, write-run-report | Added now | eval_run (grade, run report), reflex_procedure (outcome ladder, verification strength), benchmark_suite (run checks) | verify_gate, compile_dag | write-handoff for each shortfall |
| Look up before generating | [Retrieval triggers and reuse](../packs/retrieval-triggers-and-reuse/README.md): look-up-first | Added now | eval_run (lookup), reflex_procedure (triggers, ladder, fingerprints, evidence types) | situate | try-reuse, record-saved-need, survey-existing |
| Reuse only under the check | Retrieval triggers and reuse: try-reuse, qualify-solution, review-promotion (a different worker or a person) | Added now | code_node (qualified solution), eval_run (reuse attempt, promotion), benchmark_suite (reuse checks) | verify_gate, distill_reflex | record-finding for a wrongly verified solution |
| Unmet needs met by later arrivals, and whether reuse pays | Retrieval triggers and reuse: record-saved-need, match-arrivals, measure-reuse | Added now | seed_hypothesis (saved need), eval_run (arrival match, reuse ledger) | generate_seed, probe_source, benchmark_harness | try-reuse per recommendation |
| A read-only inventory, placements and proven duplicates | [File triage at scale](../packs/file-triage-at-scale/README.md): scope-scan, inventory-files, classify-items, find-duplicates | Added now | seed_hypothesis (scope, register), eval_run (inventory, duplicate set), reflex_procedure (placement, duplicates, filesystem fit, secret exclusions) | problemize, ingest, situate, verify_gate | plan-moves |
| A move plan a person approves, and verified moves | File triage at scale: plan-moves, approve-plan (a person's verb), move-and-verify, retire-sources (a person's verb), reconcile-indexes | Added now | seed_hypothesis (move plan), eval_run (approval, journal, retirement, reconciliation), benchmark_suite (triage checks) | decompose_dag, verify_gate (human), compile_dag, recheck_fragility | record-finding for drift |
| Claims that say how they are known | [Measured claims](../packs/measured-claims/README.md): label-claims, reproduce-claim, record-correction, audit-note (a different worker or a person) | Added now | seed_hypothesis (note), eval_run (reproduction, correction, audit), benchmark_suite (claim checks) | situate, benchmark_harness, recheck_fragility, verify_gate | reproduce-claim for each reported number a conclusion needs |

**Still needed here:** turning one request into many varied questions, personas and situations for
testing; and lineage and budget-derived grants for one task split across many workers over days. A
research dossier kept current in cycles, a panel of independent reviewers whose disagreements are
compared, and wiring checks that fail when a capability is reachable from no live path are now
packs, in section 9.

## 7. Testing, acceptance and accessibility

Added October 30, 2026, so a surface can be tested the way the people using it would, audited
against WCAG 2.2 AA, explored for what no script asked about and checked before anyone shares its
link. Every pack keeps the decision with a person: a sign-off, an assistive-technology session or a
go decision is a person's verb. They sit beside section 4's screenshot design review, which measures
visual design rules from captures: the accessibility audit cites WCAG criteria and the heuristics
review judges whole tasks.

| Step | Pack and verb | Status | Noun kinds | Verb kinds | Proposes |
| --- | --- | --- | --- | --- | --- |
| Personas and criteria a person agrees to before testing | [User acceptance testing](../packs/user-acceptance-testing/README.md): draft-personas, write-acceptance-criteria, agree-criteria (a person's verb) | Added now | seed_hypothesis (personas, criteria), eval_run (agreement), skill_pack (guides) | solve_draft, problemize, verify_gate (human) | write-scenario-scripts |
| Scripts run on a named build, results recorded as observed | User acceptance testing: write-scenario-scripts, run-scenario | Added now | benchmark_suite (script), eval_run (session record), reflex_procedure (defect and test-data rules) | solve_draft, benchmark_harness | record-finding per failed or blocked step |
| Verdicts by a fixed rule and a person's go or no-go | User acceptance testing: compile-results, sign-off (a person's verb) | Added now | eval_run (summary, sign-off), benchmark_suite (sign-off checks) | verify_gate | |
| A WCAG 2.2 AA audit, every finding citing its criterion | [Accessibility audit](../packs/accessibility-audit/README.md): plan-audit, capture-states, check-contrast, lint-structure, walk-keyboard, check-reflow, check-target-size, check-media, check-motion, check-text-alternatives | Added now; the contrast and lint code is tested | seed_hypothesis (plan), eval_run (captures, check results), benchmark_suite (checklist, audit checks), public_source (WCAG 2.2), code_node (contrast and lint tool) | problemize, ingest, verify_gate | record-finding per failure |
| A person with assistive technology, the report and its review | Accessibility audit: check-with-assistive-tech (a person's verb), write-audit-report, review-audit (a different worker) | Added now | eval_run (session, report, review) | verify_gate, solve_draft | verify-fix after fixes |
| Ten heuristics, independent evaluators and a fix verb per finding | [Usability heuristics review](../packs/usability-heuristics-review/README.md): scope-review, evaluate-against-heuristics, combine-findings, review-findings (a different worker), propose-fixes | Added now | skill_pack (ten heuristics), reflex_procedure (severity, combine and evidence rules), eval_run (sheets, combined findings, review), seed_hypothesis (scope, fix plan) | problemize, verify_gate, decompose_dag | propose-fix, draft-embodiments, plan-audit |
| Links, claims, analytics, error pages, rollback and the owner's go | [Launch readiness](../packs/launch-readiness/README.md): plan-readiness, check-live-links, check-page-claims, check-analytics, check-error-pages, review-rollback, assemble-readiness, record-go-decision (a person's verb) | Added now | benchmark_suite (checklist, go checks), reflex_procedure (claim, analytics and error page rules), eval_run (checks, report, decision) | decompose_dag, verify_gate | record-finding per mismatch, prepare-rollback |
| Time-boxed exploration with charters, debriefs and coverage | [Exploratory testing charters](../packs/exploratory-testing-charters/README.md): draft-charters, run-session, debrief-session (a different worker), map-coverage | Added now | seed_hypothesis (charter), eval_run (notes, debrief, coverage), skill_pack (charter guide, test ideas), reflex_procedure (session rules, coverage model) | generate_seed, benchmark_harness, verify_gate | record-finding per bug, draft-charters per gap |

**Still needed here:** moderated usability sessions with real participants, which need consent
and human-sensitive handling; load testing; contract tests for public APIs against their published
descriptions; and testing translated surfaces with fluent reviewers. Performance budgets kept against
an accepted baseline, and API response shapes, are now in section 10's regression watch pack.

## 8. Editing, reviewing and improving a video

Added October 30, 2026, at the owner's request for editing, review, QA and acceptance packs that can be
used to improve a video: an edit paced to its words, measured checks beyond the eight post-encode
checks, a rubric review with a fix verb per finding, cut-downs that keep their meaning, honest covers
and titles, and whether the intended viewers understand it. They hand the page and player to the accessibility
audit pack, single frames with known layout to the screenshot design review pack, and criteria
agreement and sign-off to the user acceptance testing pack, instead of redefining them.

| Step | Pack and verb | Status | Noun kinds | Verb kinds | Proposes |
| --- | --- | --- | --- | --- | --- |
| An edit decision list paced to the words | [Video editing](../packs/video-editing/README.md): draft-edit-list, check-pacing | Added now | scene_recipe (edit list), eval_run (pacing report), reflex_procedure (pacing, split edit, b-roll and transition rules) | solve_draft, verify_gate | revise-edit for flagged shots |
| Mix, colour, reframes, conform and export | Video editing: plan-mix, check-color, plan-reframes, write-export-spec, conform-edit | Added now | scene_recipe (mix plan, reframe plan, export spec), eval_run (colour report, conform record), reflex_procedure (mix, colour, reframing, conform rules) | solve_draft, verify_gate | render-sequence per profile |
| Findings answered in new versions, and a person's lock | Video editing: revise-edit, lock-cut (a person's verb) | Added now | scene_recipe (edit list), eval_run (revision note, lock record), benchmark_suite (edit checks) | solve_draft, verify_gate (human) | re-review |
| Measured checks beyond the eight post-encode checks | [Video QA checks](../packs/video-qa/README.md): collect-measurements, lint-captions, check-picture, check-sound, check-file-spec, check-legibility, compile-qa-report | Added now; the code is tested | scene_recipe (QA plan, text items), eval_run (measurement set, check rows, extended QA report), code_node (QA checks), benchmark_suite (extended checks, report checks) | compile_dag, verify_gate | record-finding per failure |
| Panels, claims and a rubric review | [Video review](../packs/video-review/README.md): make-panels, trace-claims, review-cut (a person or a different worker) | Added now | seed_hypothesis (scope), eval_run (panel set, claim trace, review findings), benchmark_suite (rubric, severity guide, review checks), reflex_procedure (fix routes) | compile_dag, verify_gate | route-findings |
| A second reviewer, a person listening, routing and re-review | Video review: compare-reviews, watch-and-listen (a person's verb), route-findings, re-review | Added now | eval_run (comparison, listening record, re-review), seed_hypothesis (fix routing) | verify_gate, verify_gate (human), decompose_dag | record-finding and one fix verb per cause |
| Cut-downs that keep their meaning | [Social cut-downs](../packs/social-cutdowns/README.md): find-hook-moments, plan-cutdown, check-meaning (a different worker), draft-copy, approve-cutdown (a person's verb) | Added now | seed_hypothesis (platform profile, copy draft), eval_run (hook candidates, meaning check, approval), scene_recipe (cut-down plan), benchmark_suite (cut-down checks) | situate, decompose_dag, verify_gate, solve_draft, verify_gate (human) | plan-reframes, compose-cut |
| Covers and titles a person picks | [Thumbnails and titles](../packs/thumbnail-and-title/README.md): draft-titles, draft-thumbnails, check-thumbnails, check-honesty (a different worker), plan-ab-test, record-ab-result (a person's verb) | Added now | seed_hypothesis (titles, A/B plan), scene_recipe (thumbnails), eval_run (size check, honesty check, A/B result), benchmark_suite (title checks, contrast example) | solve_draft, verify_gate, verify_gate (human) | assemble-comparison, record-choice |
| Whether the intended viewers understand it | [Video audience acceptance](../packs/video-uat/README.md): plan-acceptance, run-session (a person's verb), score-answers (a different worker), decide-acceptance, write-uat-findings | Added now | seed_hypothesis (plan), eval_run (session notes, scored answers, decision), benchmark_suite (task bank, acceptance checks) | problemize, verify_gate (human), verify_gate, decompose_dag | agree-criteria and sign-off in user acceptance testing, record-finding per miss |

**Still needed here:** checking each platform's current limits on the day, which the cut-down
profiles leave to a person; comparing a posted cut-down's retention with its long cut's, which needs
numbers only the platform shows a person; and the caption translation and audio description that
section 1 still lists.

## 9. Research that stays current, wiring checks and second opinions

Added October 30, 2026, from the same recurring patterns as section 6: a research dossier kept current
long after its first answers, proof that a code project's tested capabilities are reachable from the
paths that run, and independent second opinions before a decision. The dossier pins the deep research
report pack's nouns, the wiring checks record every gap in the findings to fixes shape, and the panel
hands disagreements of taste to the taste palettes pack and factual ones to the evaluation design pack.

| Step | Pack and verb | Status | Noun kinds | Verb kinds | Proposes |
| --- | --- | --- | --- | --- | --- |
| A numbered question queue for a topic that must stay current | [Living research dossier](../packs/living-research-dossier/README.md): open-dossier | Added now | seed_hypothesis (dossier), reflex_procedure (queue, cycle, decay class and trigger rules) | problemize | plan-cycle |
| Cycles with one sub-agent per question, folded in and logged | Living research dossier: plan-cycle, answer-question, fold-in, log-cycle | Added now | seed_hypothesis (cycle plan, answer, dossier), public_source (source log), eval_run (cycle log) | decompose_dag, solve_draft, situate, verify_gate | answer-question per picked question |
| Named triggers once the questions run out, and decayed facts read again | Living research dossier: switch-to-watching, check-triggers, refresh-decayed-facts | Added now | seed_hypothesis (dossier), eval_run (trigger check, refresh record) | situate, probe_source, recheck_fragility | fold-in, plan-cycle |
| A different worker's review and a person's publication decision | Living research dossier: review-dossier, decide-publication (a person's verb) | Added now | eval_run (adversarial review, delivery record), benchmark_suite (dossier checks) | verify_gate, verify_gate (human) | answer-question for a missed counter search |
| Entry points, dark modules, config round trips and probes that probe | [Wiring and conformance checks](../packs/wiring-and-conformance-checks/README.md): scope-wiring, find-unused-modules, check-config-round-trip, check-health-probes | Added now; the code is tested | seed_hypothesis (scope), eval_run (module map, round-trip report, probe report), code_node (wiring checks) | problemize, verify_gate | record-wiring-findings |
| Documented behaviors on their real path, a canary per rule and one conformance report | Wiring and conformance checks: map-behaviors-to-tests, plant-canaries, record-wiring-findings, compile-conformance-report | Added now | eval_run (behavior map, conformance report), benchmark_suite (canary set, wiring checks), seed_hypothesis (findings in the findings to fixes shape) | verify_gate, problemize, compile_dag | triage and propose-fix per finding |
| A panel of stated perspectives on other model families | [Persona panel and second opinion](../packs/persona-panel-second-opinion/README.md): define-panel | Added now | seed_hypothesis (panel), prompt_shim (perspective cards), reflex_procedure (panel, independence and synthesis rules) | problemize | give-view per reviewer |
| Takes given alone, checked before anyone reads them | Persona panel and second opinion: give-view, check-independence | Added now | eval_run (view, independence check) | verify_gate | give-view again for an excluded view |
| Disagreements first, audited, and a person's decision | Persona panel and second opinion: synthesize-disagreements, audit-synthesis, decide (a person's verb) | Added now | seed_hypothesis (synthesis, palette brief), eval_run (audit, decision), benchmark_suite (panel checks) | solve_draft, verify_gate, verify_gate (human) | design-evaluation for a factual disagreement, draft-embodiments for one of taste |

**Still needed here:** wiring checks for languages other than Python; dossier sources that sit behind a
login the requester holds, which need their own access rules; and a panel of real stakeholders instead of
perspective cards, which needs consent and human-sensitive handling.

## 10. Self-review, continuous review, annoyances, edge cases, regressions and code health

Added October 30, 2026, at the owner's request for nouns and verbs that help a project review itself
again and again and find the annoying things, annoying edges and corner cases of its videos, sites,
APIs and code. Every finding they produce is a findings to fixes finding, with reproduce, propose-fix,
fix, verify-fix by a different worker and add-regression-guard proposed. They pin the checks of the
packs that own them instead of redefining them: the codebase health review leaves unused modules,
mock-only tests and hollow probes to the wiring and conformance checks pack, the regression watch
compares the video QA pack's numbers and the release and verify pack's smoke results, and the review
rotation's approval is the unattended run record pack's run approval.

| Step | Pack and verb | Status | Noun kinds | Verb kinds | Proposes |
| --- | --- | --- | --- | --- | --- |
| Scopes with their checks, and a rotation a person approves | [Continuous self-review](../packs/continuous-self-review/README.md): map-scopes, approve-schedule (a person's verb) | Added now | seed_hypothesis (scope catalog, schedule proposal), eval_run (run approval) | problemize, verify_gate (human) | pick-scope |
| The next scopes by a fixed rule, each run diffed against the last and logged | Continuous self-review: pick-scope, diff-against-last-run, log-run | Added now | eval_run (scope pick, review run, review log), reflex_procedure (rotation and diff rules), benchmark_suite (loop checks) | recheck_fragility, verify_gate, compile_dag | the scope's checks, verify-fix, reproduce, triage |
| A worker's own change reviewed before hand-off, and what the loop misses | Continuous self-review: review-own-change, review-coverage | Added now | eval_run (self-review, coverage review) | verify_gate, distill_reflex | record-finding, validate-checks, review-patterns |
| Friction tours as charters, walked and counted, and complaints in people's words | [Annoyance hunt](../packs/annoyance-hunt/README.md): draft-friction-charters, hunt-annoyances, collect-complaints | Added now | seed_hypothesis (charters in the exploratory form), eval_run (friction session, complaint extract), benchmark_suite (friction catalog) | generate_seed, benchmark_harness, ingest | rank-annoyances |
| Frequency times pain, a person's confirmation and a finding per annoyance | Annoyance hunt: rank-annoyances, confirm-ranking (a person's verb), record-annoyance-findings | Added now | eval_run (annoyance log, confirmation), seed_hypothesis (findings), reflex_procedure (pain scale, frequency rule) | situate, verify_gate (human), problemize | propose-fix, a palette brief for a taste call |
| Expected results first, then seeded labelled cases | [Edge and corner cases](../packs/edge-and-corner-cases/README.md): plan-edges, generate-cases | Added now; the generator is tested | seed_hypothesis (edge plan), eval_run (case set), benchmark_suite (catalogs, labels), code_node (generator) | problemize, generate_seed | run-input-cases and the check verbs |
| Inputs, interface states, video edges and API edges, each failure a finding | Edge and corner cases: run-input-cases, check-interface-edges, check-video-edges, check-api-edges, record-edge-findings | Added now | eval_run (case results), seed_hypothesis (findings) | benchmark_harness, verify_gate, problemize | add-regression-guard with the failing case as the test |
| Baselines a person accepts, and every change measured again | [Regression watch](../packs/regression-watch/README.md): capture-baseline, accept-baseline (a person's verb), apply-acceptance, remeasure | Added now | eval_run (baseline set, acceptance, new baseline set, measurement), skill_pack (baseline kinds) | benchmark_harness, verify_gate (human) | compare-to-baseline |
| Mismatches rerun before reporting, regressions and flaky checks as findings | Regression watch: compare-to-baseline, report-regressions, review-flaky-checks | Added now | eval_run (comparison, flake review), seed_hypothesis (findings), reflex_procedure (compare and flake rules) | verify_gate, problemize, distill_reflex | accept-baseline, validate-checks |
| One commit, area by area, with a TODO ager and a swallowed-exception finder | [Codebase health review](../packs/codebase-health-review/README.md): plan-health-review, age-todos, find-swallowed-exceptions, check-dependencies, measure-code-structure, review-tests-and-errors, scan-secret-patterns | Added now; the ager and finder are tested | seed_hypothesis (plan), eval_run (area results), benchmark_suite (health areas), code_node (health checks) | problemize, verify_gate, recheck_fragility, situate | write-health-report |
| A report with no single score, and a finding per item that needs action | Codebase health review: write-health-report | Added now | eval_run (health report), seed_hypothesis (findings) | compile_dag | propose-fix, check-deprecations, find-unused-modules |

**Still needed here:** moderated sessions with real people for the annoyance hunt, which need consent and
human-sensitive handling; health checks for languages other than Python and for coverage tools' own
output formats; and pixel-level comparison of screenshots that differ only in antialiasing, which the
regression watch leaves to a pinned capture setup and a person.

## 11. Video composition QA, harness feeds, ecosystem search, LoRA shimming and statistical hygiene

Added October 30, 2026, with tested code in every pack: frame-level UI occlusion and audio-narration
QA alongside 3D perspective camera trajectories, encapsulated open agent harness feeds, task and
task-object ecosystem indexing, post-cutoff LoRA vs shim routing, token-budget context knapsack
packing and non-advisory statistical validation hygiene.

| Step | Pack and verb | Status | Noun kinds | Verb kinds | Proposes |
| --- | --- | --- | --- | --- | --- |
| Every-frame detectors plus UI mask occlusion checks | [Frame-by-frame video QA](../packs/frame-by-frame-video-qa/README.md): sample-frames, run-detectors, segment-ui-masks, cluster-and-report, draw-review-sample, label-review-sample (human), compute-precision | Added now; the code is tested | benchmark_suite (spec), reflex_procedure (safe areas), code_node (frame QA), scene_recipe (templates), eval_run (examples) | ingest, verify_gate, compile_dag | fix verbs per confirmed finding |
| Loudness, sidechain pumping, edit clicks, speaking WPM and ASR CER | [Audio and narration QA](../packs/audio-and-narration-qa/README.md): measure-loudness-and-peaks, audit-ducking-and-clicks, check-pacing-and-caption-sync, gate-narration-asr-cer, compile-narration-qa-report | Added now; the code is tested | benchmark_suite (spec), reflex_procedure (rules), code_node (audio QA), scene_recipe (report schema), eval_run (examples) | verify_gate, compile_dag | narration re-takes or mix fixes |
| 1x3 and 2x2 3D hub trajectories, safe bands and zero double-exposure cuts | [Parametric 3D video composer](../packs/parametric-3d-video-composer/README.md): plan-hub-camera-trajectory, layout-browser-and-verify-safe-band, validate-transitions-and-curves, compile-composition-check | Added now; the code is tested | motion_profile (camera spec), reflex_procedure (layout rules), code_node (composer), scene_recipe (check schema), eval_run (examples) | solve_draft, verify_gate, compile_dag | stage render tasks |
| Encapsulated Agent Harnesses Feed with HarnessTax and Pareto selection | [Agent harness feeds](../packs/agent-harness-feeds/README.md): probe-harness-sources, normalize-harness-benchmarks, recheck-harness-freshness, encapsulate-harness-feed, select-harness-for-task | Added now; the code is tested | feed_source (encapsulating feed), model_spec (harness profiles), public_source, eval_run (matrix), seed_hypothesis (schedule, Pareto), code_node | fetch_feed, benchmark_harness, recheck_fragility, verify_gate, situate | normalize-harness-benchmarks on drift |
| Repository metadata decomposed into searchable tasks and task objects | [Hackathon ecosystem index](../packs/hackathon-ecosystem-index/README.md): decompose-repo-catalog, search-hybrid-rrf, verify-catalog-hygiene | Added now; the code is tested | benchmark_suite (spec), reflex_procedure (decomposition rules), code_node (indexer), scene_recipe (schemas), eval_run (examples) | ingest, situate, verify_gate | prior-art surveys |
| Post-cutoff LoRA pairs, shim routing and paired bootstrap gates | [LoRA memory and shimming](../packs/lora-memory-and-shimming/README.md): route-shim-or-lora, compile-lora-pairs, evaluate-bootstrap-gate | Added now; the code is tested | benchmark_suite (gates), reflex_procedure (pair rules), code_node (compiler/router), scene_recipe (gate schema), eval_run (examples) | situate, distill_reflex, verify_gate | holdout refresh |
| Hybrid BM25 + dense RRF, MinHash dedup, injection stripping and 0/1 knapsack | [Hybrid retrieval context knapsack](../packs/hybrid-retrieval-context-knapsack/README.md): fuse-hybrid-rrf, sanitize-and-dedup-chunks, pack-token-knapsack | Added now; the code is tested | benchmark_suite (RRF/knapsack spec), reflex_procedure (sanitization), code_node (packer), scene_recipe (receipt), eval_run (examples) | situate, verify_gate, compile_dag | corpus cleanup |
| Non-advisory purged walk-forward splits, regime Rank IC, turnover and DSR | [Leakage-safe regime backtesting](../packs/leakage-safe-regime-backtesting/README.md): build-purged-splits, score-regime-ic-and-turnover, deflate-multiple-testing | Added now; the code is tested | benchmark_suite (hygiene spec), reflex_procedure (formulas), code_node (auditor), scene_recipe (receipt), eval_run (synthetic panel) | compile_dag, benchmark_harness, verify_gate | feature leakage fixes |
| Multi-frame dwell, static-slide zoom lock, split-attention load, desktop web journey provenance and frosted captions | [Scene and cognitive video QA](../packs/scene-and-cognitive-video-qa/README.md): audit-shot-dwell-and-transitions, audit-camera-and-slide-stability, audit-reading-and-cognitive-load, audit-web-journey-and-captions, compile-scene-cognitive-report | Added now; the code is tested | benchmark_suite (gates), reflex_procedure (frosted caption rule), code_node (multi-frame QA), scene_recipe (report schema), eval_run (worked example) | verify_gate | scene re-edits, static camera locks and caption re-renders |
| 1,000x1,000x1,000x1,000 dynamic scenario generation, stakeholder boundary gating and candidate task and task-object digestion | [Meta ARE scenario matrix](../packs/meta-are-scenario-matrix/README.md): generate-four-axis-catalog, evaluate-stakeholder-episodes, digest-reusable-nouns-and-verbs | Added now; the code is tested | benchmark_suite (4-axis matrix), reflex_procedure (stakeholder rules), code_node (matrix generator), scene_recipe (digest schema), eval_run (worked example) | generate_seed, verify_gate, distill_reflex | candidate tasks (things to do) and task objects for distinct review |
| Offline SIRA pack enrichment, RA-RFT structural analogy retrieval, sub-2ms shim compilation and DGM-H distinct verification | [SIRA self-learning RAG](../packs/sira-self-learning-rag/README.md): enrich-corpus-and-analogy-frames, compile-sub2ms-shim-rules, verify-dgmh-proposal-independently | Added now; the code is tested | benchmark_suite (SIRA/RA-RFT spec), reflex_procedure (gates), code_node (enricher/indexer/verifier), scene_recipe (receipt schema), eval_run (worked example) | generate_seed, distill_reflex, verify_gate | owner review of verified DGM-H proposals |
| 5-tier compute routing across sub-2ms edge shims, Fireworks multi-LoRA rubrics, Baseten 2M TPM swarms and Modal QLoRA/sandboxes | [Partner compute cascade](../packs/partner-compute-cascade/README.md): validate-modal-qlora-and-sandbox, pace-baseten-fenced-swarm, route-fireworks-multilora-and-rubric | Added now; the code is tested | model_spec (5-tier cascade), reflex_procedure (gates), code_node (validators/clients), scene_recipe (receipt schema), eval_run (worked example) | verify_gate, benchmark_harness | escalated Baseten or Modal tasks on failed specialist rubric |
| Combinatorial multi-domain niche cells, BFS spider frontier, refresh movers/404s and 8-axis yield benchmarks | [Combinatorial niche spider](../packs/combinatorial-niche-spider/README.md): mint-niche-cells-and-paths, sweep-cells-and-spider-frontier, review-eight-axis-admissions | Added now; the code is tested | benchmark_suite (8-axis gates), reflex_procedure (cursor/spider rules), code_node (forge/spider/evaluator), scene_recipe (receipt schema), eval_run (worked example) | generate_seed, distill_reflex, verify_gate | bounded Wave 1-3 follow-up verb proposals and curated admissions |
| Enforcing scoped effect ceilings, credential and high-entropy secret redaction, and model data-tier routing guards that block non-public payloads from training or contributor endpoints before dispatch, with tested code | [Agent capability and tier firewall](../packs/agent-capability-and-tier-firewall/README.md): author-capability-tier-policy, scan-payload-for-secrets, evaluate-pre-dispatch-gate, quarantine-blocked-dispatch | Added now; the code is tested | benchmark_suite, reflex_procedure, code_node, scene_recipe, eval_run | generate_seed, verify_gate | distinct review of candidate outputs |
| Scanning repository agent definitions, rule files, MCP configs, and self-editing memory files for hidden Trojan Unicode, unapproved hook or server escalation, built-in name shadowing, and authority spoofing, with tested standard-library Python code, with tested code | [Agent config supply chain scanner](../packs/agent-config-supply-chain-scanner/README.md): scan-repo-agent-configs, gate-self-editing-memory-diff, quarantine-supply-chain-finding | Added now; the code is tested | benchmark_suite, reflex_procedure, code_node, scene_recipe, eval_run | generate_seed, verify_gate | distinct review of candidate outputs |
| Deduplicating noisy raw notes, chunking unstructured text into content-addressed records, and clustering real-world situations into candidate seed-pack categories, with tested standard-library Python code, with tested code | [Brainstorm chunking and situation miner](../packs/brainstorm-chunking-and-situation-miner/README.md): deduplicate-raw-notes, chunk-canonical-notes, extract-real-world-situations, score-category-coverage-gaps | Added now; the code is tested | benchmark_suite, reflex_procedure, code_node, scene_recipe, eval_run | generate_seed, verify_gate | distinct review of candidate outputs |
| Mining sanitized multi-hour AI coding session JSONL transcripts into reusable failure tombstones, tool-error loops, verified fix recipes, and prompt-shim candidates, with tested standard-library Python code, with tested code | [Coding session trace distiller](../packs/coding-session-trace-distiller/README.md): redact-session-trace, mine-error-loops-and-tombstones, extract-verified-fix-recipes, draft-prompt-shim-candidates, compile-distillation-report | Added now; the code is tested | benchmark_suite, reflex_procedure, code_node, scene_recipe, eval_run | generate_seed, verify_gate | distinct review of candidate outputs |
| Declarative multi-dimensional configuration grids, runtime parameter clamping, verified cell lifecycles, contextual regime selection, saturated boundary auto-expansion and AST magic-literal gates, with tested code | [Configuration grids and bounds](../packs/configuration-grids-and-bounds/README.md): enumerate-and-clamp-grid, scan-zero-magic-literals-ast, verify-and-promote-regime-cell | Added now; the code is tested | benchmark_suite, reflex_procedure, code_node, scene_recipe, eval_run | generate_seed, verify_gate | distinct review of candidate outputs |
| Compiling structured data tables and component message traces into publication-grade accessible SVGs (confidence-interval forest plots, progression bars, and edge-packet diagrams) with deterministic WCAG contrast and label-collision checks, with tested code | [Declarative chart and diagram SVG](../packs/declarative-chart-and-diagram-svg/README.md): specify-declarative-figure, compile-accessible-svg, audit-svg-contrast-and-collisions, revise-colliding-or-faint-layout | Added now; the code is tested | benchmark_suite, reflex_procedure, code_node, scene_recipe, eval_run | generate_seed, verify_gate | distinct review of candidate outputs |
| Compiling four-stage portable execution hooks across coding harnesses, enforcing pre-tool safety and spend guards, running topological pre-submit verifier DAGs, auditing verifier mutation sensitivity and extracting dead-end negative tombstones, with tested code | [Execution hooks and verifiers](../packs/execution-hooks-and-verifiers/README.md): compile-portable-hooks-and-guard-tools, execute-verifier-dag-and-tombstones, audit-verifier-mutation-sensitivity | Added now; the code is tested | benchmark_suite, reflex_procedure, code_node, scene_recipe, eval_run | generate_seed, verify_gate | distinct review of candidate outputs |
| Evaluating declarative film and surface rule nouns, blocking forbidden synthetic or non-functional stamps, scoring frosted-pill caption centering and dual-channel cognitive load, repointing dead repository links to SHA-256-pinned source mirrors and verifying llms.txt parity against MCP tools, with tested code | [Film and surface rule evaluator](../packs/film-and-surface-rule-evaluator/README.md): evaluate-film-rules-and-caption-load, repoint-dead-links-and-check-mcp-parity, review-film-and-surface-readiness | Added now; the code is tested | benchmark_suite, reflex_procedure, code_node, scene_recipe, eval_run | generate_seed, verify_gate | distinct review of candidate outputs |
| Aho-Corasick literal scanning, portable grep rules, Unicode and entropy security checks and strict JSON and copy formatting, with tested code | [Grep, string and format rules](../packs/grep-string-and-format-rules/README.md): scan-literals-and-portable-grep-rules, audit-unicode-smuggling-and-entropy, verify-strict-json-copy-and-ast-gate | Added now; the code is tested | benchmark_suite, reflex_procedure, code_node, scene_recipe, eval_run | generate_seed, verify_gate | distinct review of candidate outputs |
| Inert window-keyed schedule compilation, fragility half-life urgency ranking, AIMD claim pacing, UCB1 anti-stall bandit loops, seven-tier compute resource cascades and six advice-only swappable learners, with tested code | [Loops schedules and heuristics](../packs/loops-schedules-and-heuristics/README.md): compile-inert-schedule-and-aimd-pacing, step-bandit-anti-stall-and-compute-cascade, evaluate-swappable-learner-advisors | Added now; the code is tested | benchmark_suite, reflex_procedure, code_node, scene_recipe, eval_run | generate_seed, verify_gate | distinct review of candidate outputs |
| Aligning image segmentation masks, speech-to-text word timings, and layout safe zones across multimodal media pipelines, with tested standard-library Python code, with tested code | [Multimodal mask and transcript alignment](../packs/multimodal-mask-and-transcript-alignment/README.md): normalize-masks-and-utterances, audit-audiovisual-cue-drift, route-non-occluding-safe-zones | Added now; the code is tested | benchmark_suite, reflex_procedure, code_node, scene_recipe, eval_run | generate_seed, verify_gate | distinct review of candidate outputs |
| Compiling portable agent role definitions across coding harnesses, scoring delegation descriptions, separating routing from permissions, and stripping accuracy-degrading persona fluff, with tested standard-library Python code, with tested code | [Portable agent persona compiler](../packs/portable-agent-persona-compiler/README.md): strip-degrading-persona-fluff, score-and-verify-role-contract, compile-multi-harness-role-bundle | Added now; the code is tested | benchmark_suite, reflex_procedure, code_node, scene_recipe, eval_run | generate_seed, verify_gate | distinct review of candidate outputs |
| Cache-line blocked Bloom filters, deletable Cuckoo filters, static Xor filters, multi-probe LSH and streaming sketches, with tested code | [Probabilistic filters and sketches](../packs/probabilistic-filters-and-sketches/README.md): build-and-query-membership-filters, expand-multiprobe-lsh-and-decontaminate, verify-sketch-bounds-and-ast-conformance | Added now; the code is tested | benchmark_suite, reflex_procedure, code_node, scene_recipe, eval_run | generate_seed, verify_gate | distinct review of candidate outputs |
| Planning vertical 9:16 and square 1:1 video cutdowns from 16:9 demos with subject-safe crop trajectories, hook-beat scoring, caption-band reflow, and claim-preservation gates, backed by standard-library Python code, with tested code | [Social video reframe and hooks](../packs/social-video-reframe-and-hooks/README.md): plan-subject-crop-trajectory, score-opening-hook-beats, reflow-safe-zone-captions, audit-claim-preservation, compile-cutdown-plan | Added now; the code is tested | benchmark_suite, reflex_procedure, code_node, scene_recipe, eval_run | generate_seed, verify_gate | distinct review of candidate outputs |
| Auditing engineering repositories against weighted evaluation rubrics covering problem specificity, verified open-weight and multimodal model usage, modular source and CI depth, and live deployment plus video walkthrough completeness, with tested standard-library Python code, with tested code | [Submission rubric and readiness audit](../packs/submission-rubric-and-readiness-audit/README.md): inventory-repository-evidence, verify-model-and-depth-signals, score-weighted-readiness-rubric | Added now; the code is tested | benchmark_suite, reflex_procedure, code_node, scene_recipe, eval_run | generate_seed, verify_gate | distinct review of candidate outputs |
| Three-tier trigger compilation with Role-Filler Frame isomorphism, conversation-aware submodular context-shim knapsacks, step-slicing HarnessTax break-even routing and privacy-safe compatibility matching, with tested code | [Trigger and shim rules](../packs/trigger-and-shim-rules/README.md): evaluate-three-tier-triggers, pack-dedup-shim-knapsack, route-regime-and-match-which-fits | Added now; the code is tested | benchmark_suite, reflex_procedure, code_node, scene_recipe, eval_run | generate_seed, verify_gate | distinct review of candidate outputs |
| Lossless dynamic byte-entropy patching and surprise-proportional compute allocation, with tested code | [Byte latent patch and entropy chunker](../packs/byte-latent-patch-and-entropy-chunker/README.md): segment-dynamic-byte-entropy-patches, allocate-surprise-proportional-patch-budget, verify-lossless-patching-and-ast-gate | Added now; the code is tested | benchmark_suite, reflex_procedure, code_node, scene_recipe, eval_run | generate_seed, verify_gate | distinct review of candidate outputs |
| Adaptive reasoning budget routing, latent state checkpoint compression and overthinking loop detection, with tested code | [Coconut latent and thought budget routing](../packs/coconut-latent-and-thought-budget-routing/README.md): route-adaptive-reasoning-budget, compress-latent-checkpoint-and-detect-loops, verify-invariant-recall-and-ast-gate | Added now; the code is tested | benchmark_suite, reflex_procedure, code_node, scene_recipe, eval_run | generate_seed, verify_gate | distinct review of candidate outputs |
| Walsh-Hadamard outlier rotation quantization, weight-decomposed adapter analysis and subspace interference merge routing, with tested code | [DoRA, SpinQuant and adapter merging](../packs/dora-spinquant-and-adapter-merging/README.md): rotate-outliers-and-quantize-weights, decompose-dora-and-route-adapter-subspaces, verify-quantization-merge-and-ast-gate | Added now; the code is tested | benchmark_suite, reflex_procedure, code_node, scene_recipe, eval_run | generate_seed, verify_gate | distinct review of candidate outputs |
| K-means++ Voronoi inverted-file partitioning, residual uint8 product quantization and exact top-R reranking, with tested code | [FAISS IVF-PQ and ANN index cascade](../packs/faiss-ivf-pq-and-ann-index-cascade/README.md): fit-ivf-pq-voronoi-and-residual-codebooks, benchmark-adc-and-exact-rerank-regimes, verify-ivf-pq-compression-and-recall-receipt | Added now; the code is tested | benchmark_suite, reflex_procedure, code_node, scene_recipe, eval_run | generate_seed, verify_gate | distinct review of candidate outputs |
| Action-conditioned latent world-state prediction, epistemic surprise energy anomaly gating and counterfactual rollout planning, with tested code | [JEPA world-state and action simulator](../packs/jepa-world-state-and-action-simulator/README.md): plan-counterfactual-rollout-trajectories, gate-step-by-epistemic-surprise-energy, verify-world-state-simulator-receipt | Added now; the code is tested | benchmark_suite, reflex_procedure, code_node, scene_recipe, eval_run | generate_seed, verify_gate | distinct review of candidate outputs |
| Monotone Rule of Two capability lattices across tool DAGs, bidirectional privacy placeholder vaults and model license-tier firewalls, with tested code | [Rule of Two and agent guardrails](../packs/rule-of-two-and-agent-guardrails/README.md): evaluate-rule-of-two-tool-dag, scrub-and-route-privacy-license-tier, verify-bidirectional-vault-and-ast-gate | Added now; the code is tested | benchmark_suite, reflex_procedure, code_node, scene_recipe, eval_run | generate_seed, verify_gate | distinct review of candidate outputs |
| Mutation-guided reviewer calibration across six seeded defect classes, proof-carrying review verification and Wilson/bootstrap CIs, with tested code | [Seeded defect and proof-carrying review](../packs/seeded-defect-and-proof-carrying-review/README.md): generate-seeded-defect-mutant-twins, verify-proof-carrying-review-verdicts, audit-wilson-and-paired-bootstrap-intervals | Added now; the code is tested | benchmark_suite, reflex_procedure, code_node, scene_recipe, eval_run | generate_seed, verify_gate | distinct review of candidate outputs |
| Iterative RPO preference curation, rubric-judge calibration gating and pre-registered holdout flywheel promotion, with tested code | [Self-rewarding and iterative DPO flywheel](../packs/self-rewarding-and-iterative-dpo-flywheel/README.md): calibrate-rubric-judge-against-verifiers, curate-iterative-rpo-preference-pairs, verify-preregistered-flywheel-promotion | Added now; the code is tested | benchmark_suite, reflex_procedure, code_node, scene_recipe, eval_run | generate_seed, verify_gate | distinct review of candidate outputs |
| Stateless contract-hydrated worker nodes, per-step state profiling, <6h keyword hydration, five-rung deterministic-to-CLI-harness waterfalls, multi-persona vector scoring and capability-gap spinout proposals, with tested code | [Self-constructing node and waterfall](../packs/self-constructing-node-and-waterfall/README.md): interrogate-and-profile-envelope, execute-weighted-waterfall-and-harness-grid, evaluate-persona-vectors-and-spinout-gap | Added now; the code is tested | benchmark_suite, reflex_procedure, code_node, scene_recipe, eval_run | generate_seed, verify_gate | distinct review of candidate outputs |
| Two-line frosted caption pill geometry, short-gap bridging, contrast checks and deterministic SVG rendering in a 1920x1080 lower band, with tested code | [Frosted widescreen captions](../packs/frosted-widescreen-captions/README.md): layout-and-bridge-caption-cues, verify-caption-band-and-contrast, render-frosted-caption-svg | Added now; the code is tested | benchmark_suite, reflex_procedure, code_node, scene_recipe, eval_run | solve_draft, verify_gate, compile_dag | distinct review of candidate outputs |
| Verifying partner compute receipts and formatting truthful local_verified vs live_api telemetry cards with honest speed-ratio attribution, with tested code | [Partner compute telemetry cards](../packs/partner-compute-telemetry-cards/README.md): audit-partner-compute-receipts, compile-telemetry-card-model | Added now; the code is tested | benchmark_suite, reflex_procedure, code_node, scene_recipe, eval_run | verify_gate, compile_dag | distinct review of candidate outputs |
| Append-only request lifecycle event reduction into plain-language status timelines, replay schedules and identifier-free requester views, with tested code | [Request status UI](../packs/request-status-ui/README.md): map-and-validate-request-events, compile-plain-status-and-replay-schedule | Added now; the code is tested | benchmark_suite, reflex_procedure, code_node, scene_recipe, eval_run | ingest, verify_gate | distinct review of candidate outputs |
| Widescreen walkthrough storyboard auditing against footage, anchor and pacing gates and headless viewport capture plan generation, with tested code | [Website-first walkthrough capture](../packs/website-first-walkthrough-capture/README.md): audit-walkthrough-storyboard-v6, emit-viewport-capture-plan | Added now; the code is tested | benchmark_suite, reflex_procedure, code_node, scene_recipe, eval_run | verify_gate, compile_dag | distinct review of candidate outputs |

## Work that supports every goal

[Repository review](../packs/repo-review/README.md), [Decomposition patterns](../packs/decomposition-patterns/README.md),
[Taste palettes](../packs/taste-palettes/README.md), [Harness onboarding](../packs/harness-onboarding-e2e/README.md),
[Cost optimization](../packs/cost-optimization/README.md) and [Docs improvement](../packs/docs-improvement/README.md)
exist in this library. Decomposition patterns is where verbs most clearly create verbs:
a breakdown becomes child task previews, and the requester enqueues them.
[Findings to fixes](../packs/findings-to-fixes/README.md), added later on October 29, turns every
error, issue, design flaw or annoyance noticed in any of this work into recorded, fixed and
verified work.

## Not packs

These are infrastructure, not seed packs, and are not built here: Tessera's flag-gated
schedule support that the daily proposals need; any importer that writes a pack's nouns;
and delivery of results to people. Every schedule noun in this library says it is not
activated.