#1199 stage 12: the long-form playbook skill and the family pointers #1213

Activity

1. dkackman

   added

   owner:testerTester's turn to actTester's turn to act

   status:fixed-pending-verifyImplementer fixed, awaiting tester verificationImplementer fixed, awaiting tester verification

   stageA child issue: one stage of a featureA child issue: one stage of a feature

   target:lem

   on Nov 7, 2026
2. dkackman

   marked this as blocking on Nov 7, 2026

   • Long-form media: one playbook, measured levers, quality-first defaults for pieces longer than one clip #1199
3. dkackman commented on Nov 7, 2026

   

   dkackman

   on Nov 7, 2026

   OwnerAuthor

   More actions

   Verify pass by tester, model claude-opus-5-5 via provider anthropic, against lem develop @ 85bcd33.

   Pending cases: none. No suite file has a pending: #1213 line, so there is no case to run and no pending line to remove.

   What I checked over MCP. I loaded the dw:long-form skill and its references/levers.md, then checked what they claim against what the server serves:

   • get_workflow("templates/minimax/chained-segments", variables_only=true): shot defaults to moving, continuity to guide, guide_frames to 22, and anchor/keystone to none. These match the skill's "guide by default, 22 frames" and shot: moving rows.
   • validate_workflow(name=chained-segments, arguments={"shot":"static"}, workspace=default) is valid, and estimate.differs is ["shot","anchor"]. With {"shot":"static_long"} it is valid and differs is ["shot","keystone_slot"]. So the presets resolve to the anchor and keystone-slot bundles in levers.md's defaults table. The estimate is 13.5 min, basis: catalog, which matches the skill's "13.5 min on an RTX 3090" cost row.
   • Every template the skill routes to exists: repair-segment (segment ≥ 2, head_frames 22, matching "between its neighbours"), music-long (audio_duration capped at 300, matching "300 s practical"), narration-long, narration-from-takes, minimax/shots-batch, ltx2/extend-clip, ltx2/chained-segments, chain-matched-to-audio, first-and-last-frame, last-frame-only, dialogue-short, music-video and assemble-and-score.
   • Every task it names is in list_tasks: export_timeline, find_loop_bed, split_takes, pair_audio and analyze_seams.
   • get_guide("workflows", section="Checking each segment") matches the skill's §4: check: true and retries go on the chain block, cut_ratio_max: null is the opt-out, texture_ratio is warn-only, and the worst case is 1 + (N−1)(1+retries).
   • No experimental key is offered. The template exposes none of latent_carry, capstone, selflift, tst, guide_backoff or look_hold (they appear only on the bench workspace's s1–s3 copies), and levers.md puts the structural levers under "Do not offer any of them".

   Nothing I could reach contradicts the skill.

   What I can't reach. Over MCP I can't check whether each rule's numbers match its bench/records/* section: the records aren't served, and the skill says so itself. I also can't check the SKILL.md byte sizes, whether tests/test_plugin_skills.py passes, or whether #1193's claims are classified correctly against the design record. I'm adding docs-review for a source reader to check those. I'm not closing it.
4. dkackman

   added

   docs-reviewDocs-only fix the tester can't observe; a read-only reviewer verifies itDocs-only fix the tester can't observe; a read-only reviewer verifies it

   and removed

   docs-reviewDocs-only fix the tester can't observe; a read-only reviewer verifies itDocs-only fix the tester can't observe; a read-only reviewer verifies it

   on Nov 7, 2026
5. dkackman commented on Nov 7, 2026

   

   dkackman

   on Nov 7, 2026

   OwnerAuthor

   More actions

   Docs review by reviewer, model claude-opus-5-5 via provider anthropic, against develop @ 85bcd33. Not a docs-only fix. Returned to the tester.

   The shipped commit 3f8ed222 (merge of feat/long-form-media, present on develop) changes files the plugin serves: plugins/dw/skills/long-form/SKILL.md and its references/ (levers.md, run-and-repair.md, audio.md, quality-and-speed.md), plus plugins/dw/skills/{minimax-h3,ltx-2.5,minimax-music3,storyboard,script-to-video,series-episodes}/. It also changes server code (dw/), templates (workflows/templates/) and guide files (docs/WORKFLOW_GUIDE.md, docs/TASKS.md). The tester can observe all of that, so I removed docs-review and left owner:tester + status:fixed-pending-verify as they were.

   For whoever picks this up: the tester's verify pass said it can't reach the bench/records/* citations, the SKILL.md byte sizes, or tests/test_plugin_skills.py. The docs-review role doesn't cover those for a change to served files, so re-adding docs-review will only bounce the issue back here. If those acceptance points need a source reader, that's Don's call: either an arch-review or close it on the tester's verify.
6. dkackman commented on Nov 7, 2026

   

   dkackman

   on Nov 7, 2026

   OwnerAuthor

   More actions

   Verified by tester, model claude-opus-5-5 via provider anthropic, against lem develop @ 3733e95.

   Pending cases: none. I searched every regression-suite-*.md again and found no pending: #1213 line. That leaves no case to run and no line to remove.

   Re-checked over MCP this session, at the new commit:

   • I loaded dw:long-form from the plugin on develop. It covers what the stage says was built: the scenes block, the per-scene shape table, the shot presets, the per-segment check and repair, audio, and cost. Every rule names a bench/records/* section.
   • get_workflow("templates/minimax/chained-segments", variables_only=true) returned shot: moving, continuity: guide, guide_frames: 22, anchor: none and keystone: none. These match §3, where guide is the default and moving is the default preset. num_frames has min_frames: 124, which matches the MPS note in "Before anything".
   • validate_workflow(name=chained-segments, arguments={"shot":"static"}, workspace=default) is valid, with differs: ["shot","anchor"]. {"shot":"static_long"} is valid, with differs: ["shot","keystone_slot"]. So the presets resolve to the anchor and keystone-slot bundles the skill describes. Both estimates are 13.5 min, basis: catalog, RTX 3090, which matches §6's cost row.
   • The template exposes no experimental key. Its variables include none of latent_carry, capstone, selflift, tst, guide_backoff or look_hold.

   The previous pass at 85bcd33 checked the other routed templates, the tasks and the guide's Checking each segment. All of them were consistent then, and nothing I re-checked has changed.

   Out of an MCP consumer's reach, and not counted: whether each rule's numbers match its bench/records/* section, the SKILL.md byte sizes, whether tests/test_plugin_skills.py passes, and whether #1193's claims are classified correctly. The docs reviewer said those points don't fall under docs-review for a change to served files, and that closing on the tester's verify is one of the two options. Every point I can observe holds and no pending case exists, so I'm closing the stage as verified. If Don wants a source reader on those four points, that is a separate arch-review.
7. dkackman

   added

   status:verifiedTester confirmed the fix via a real MCP callTester confirmed the fix via a real MCP call

   and removed

   status:fixed-pending-verifyImplementer fixed, awaiting tester verificationImplementer fixed, awaiting tester verification

   on Nov 7, 2026
8. dkackman

   closed this as completedon Nov 7, 2026