10 KiB

Raw Blame History

Execution Plan

Current Medium Branch

Branch: MV shot-image production for later video assembly.
Format: 16:9 horizontal.
Production mode: image assets first; video generation/editing remains manual or semi-manual until explicitly automated.
Prompt ownership: Codex owns final image prompts; GPT intake is source specification, not final prompt.

Current Execution Focus

Shot-image production is complete for the current 51-shot table. Next work is assembly / export / closeout when requested.

Immediate focus:

Do not dispatch more production image packages unless the user requests append-only repair/retry for named slides.
Use direct child sessions for normal production-light packages.
Keep review none unless the user explicitly asks for review.
Use test/audit routing only for repairs, workflow debugging, or explicit investigation.

Materialization Targets

Target	Required Output	Status
Character anchor board	Main 16:9 board on shallow gray-white background	accepted v2
Character transparent slices	Front full body, side full body, back full body, walking side, walking back	accepted v2-from-anchor set
Character labels	Short identity, clothing, prop, and boundary labels	pending
Style anchor board	16:9 dark-gray board with A/B/C/D samples	accepted
C style retest	Dark muddy puddle / broken star reflection sample	accepted
Shot test package	S-01, S-10, S-46, S-47, S-51	accepted

Small-Batch Strategy

Do not generate all 51 shots first.

Round 1 uses five shots because they test the system's biggest risks:

Shot	Why It Is In Round 1	Anchor
S-01	Opening landscape: proves world-before-person and main A composition.	A
S-10	Steel canyon: proves city pressure without tech/UI drift.	B
S-46	Muddy starlight: proves the soul image and C repair.	C
S-47	Night road fire: proves restrained near-fire transition after C.	D
S-51	Final wide: proves distant fire, no victory, no gathering, person remains small.	D

Round 1 success means these are true:

The traveler stays recognizable without becoming a portrait subject.
A/B/C/D feel like one MV, not four unrelated styles.
C is no longer sunrise/pretty reflection and has clear muddy starlight.
D is warm but not camping, gathering, or victory.
The output can support later prompt batching without rewriting the visual system.

Current Round

Round: R2 five-shot trial preparation.
R0 completed: intake cleaned; project execution files initialized.
R1 completed so far: v2 main character anchor accepted; identity lock approved for slice generation; five transparent slice candidates generated and rejected due identity drift from v2.
R1 completed: v2 main character anchor accepted; v2-from-anchor transparent slice set accepted.
R1 completed: C style retest accepted as style-anchor-c-repair-v1.png.
R1 completed: common style anchor board accepted as style-anchor-board-v1.png.
R2 completed: S-01 generated once and accepted by user.
R2 completed: user waived intermediate prompt review for S-10; S-10 generated once and accepted by user.
R2 completed: S-46 generated once and accepted by user.
R2 completed: S-47 generated once and accepted by user.
R2 completed: S-51 generated once and accepted by user.
R2 completed: five-shot visual trial generated and accepted by user.
R3 character-consistency workflow test completed: s21-v3, s14-v2, and s25-v2 generated through the Advisor / visible-reference path and accepted by user. Quality route is usable; duplicate Agent S-14 remains a historical mechanism defect from this run.
R3 third-round workflow test completed: session-04 generated s33-v2 and s06-v2 through a clean user-visible child session plus internal one-shot SubAgents. Both images were accepted by user; review files were absent as requested; no duplicate-worker defect was observed in this two-slide scope.
R4 completed visually with mechanism warnings: clean child session 019ef8cf-4ab1-7b41-877b-979dc25b0a76 generated s02-v2, s03-v2, and s04-v2; user accepted all three images and review files were absent as requested. Mechanism is not a clean pass because brief/packet files were written by slide Agents rather than prewritten by the visible child session; S-04 also has a worker-launch sequencing warning.
R5 brief-ownership test completed: clean child session 019ef907-16b7-7a43-9924-227892aa62d3 generated accepted s07-v2, s08-v3, and s09-v2 with child-owned brief/packet before worker launch for successful outputs. s08-v2 remains incomplete transport-failure evidence and was retried append-only as s08-v3.
Package-size trials completed through a user-authorized 4-slide package. session-07 produced accepted s11-v2, s12-v2, and s13-v2 with observed context 157K and elapsed time 30m5s. session-08 mechanically completed s15-v2 through s18-v2, but s16-v2 failed visual acceptance because it did not achieve 横向大远景，路从画面一侧切向远方，人物极小 and was too similar to s15.
R6 Phase 1 notification/wakeup test completed and user accepted the image: clean child session 019ef9bc-fe96-75b3-b902-5d5ef95feca3 generated append-only s16-v3 and sent a callback to parent thread 019ef9b5-cc39-7a42-91b1-f9b0852dc95e; parent post-processing used local handoff/file evidence without polling the child thread.
R7 production-light simplification test completed: session-12 generated s26-v2, s27-v2, and s28-v2 with review none; user accepted all three images and observed 25m48s / 134K tokens.
R7 direct-child production-light test completed: session-13 generated s29-v2, s30-v2, and s31-v2 directly without Agent/SubAgent; user accepted all three images and observed 10m32s / 114K tokens. This is now the formal production default.
Production capacity validation completed: direct child session 019efc24-96a4-7582-bb05-914b82a726d5 generated s32-v2, s34-v2, s35-v1, s36-v1, and s37-v1 without Agent/SubAgent or review files; user accepted all five images. User observed callback at about 152K tokens / 58% context and final usage after post-callback document/check work at about 164K tokens / 63% context.
8-slide compression validation completed: direct child session 019efc48-d4af-7972-8646-6d04778dfb2c generated s38-v1 through s45-v1 without Agent/SubAgent or review files; user promoted the package to production-pass by production-stage default. User observed 174K tokens / 67% context / 38m37s elapsed. Ordinary production-light packages may now use up to 8 slides, but child callback must be the final action.
Final production package completed: direct child session 019efc78-e0f1-7601-aaec-bab4f3e6f99d generated s48-v1, s49-v1, and s50-v1 without Agent/SubAgent or review files; child reported complete, problems: none, needs_user: none; parent verification found expected files present, exact review files absent, and no post-callback child writes.

Active Batch

No active child production package is running. R7 production-light outputs s26-v2, s27-v2, s28-v2, s29-v2, s30-v2, and s31-v2, capacity-validation outputs s32-v2, s34-v2, s35-v1, s36-v1, and s37-v1, 8-slide validation outputs s38-v1 through s45-v1, and final production outputs s48-v1 through s50-v1 are accepted into production-pass. Direct-child, no-Agent topology remains the current production route for future repair/retry packages.

No single-shot gate is active. The five-shot visual trial is accepted. The character-consistency workflow test is visually accepted for S-21, S-14, S-25, S-33, S-06, S-07, S-08, S-09, S-11, S-12, S-13, S-16/s16-v3, S-20/s20-v3, S-22/s22-v3, S-23/s23-v3, S-24/s24-v2, S-26 through S-32, and S-34 through S-50. R4 images for S-02, S-03, and S-04 are visually accepted but remain a mechanism-warning run because child-session brief ownership did not match the desired boundary. S-16/s16-v2 remains rejected historical evidence.

Current broad-run issue report:

slides/broad-generation-self-audit-v1.md

Open Decisions

Decision	Default
Character source face	Current stage does not use the real face; use a 50-year-old traveler based on the defined character system.
Character board size	16:9 page, shallow gray-white background.
Character board scope	MVP can start with front/side/back full body plus half-body state; full version should add walking side/back and prop breakdowns.
Style board scope	MVP one-page A/B/C/D board; C should be marked as repaired / accepted.
Full shot production	Authorized for production-light. Current default route: named clean user-visible child sessions (`子会话：...`) handle up to 8 ordinary production slides directly, write brief/prompt/image/slide status/handoff, use append-only version scanning with overwrite disallowed, suppress review files, skip Agent/SubAgent, skip packet/return artifacts, skip routine final completeness audit, and callback the parent as the final child action when complete or blocked. Use 1-3 slides for identity-heavy, repair-heavy, core-risk, or context-expensive scopes. MiniMax MCP/external vision remains off unless explicitly authorized.

Next Decision

Next decision:

Shot-image production is complete for the current 51-shot table; there is no remaining not-started production scope.
Keep callback-only behavior: parent dispatches and exits; child completes all writes/checks/handoff work first, then callbacks parent as the final action. Parent reads local handoff by default and inspects details only on warning/blocked/failed.
Keep the quality route: gpt-image-2 Advisor prompt/spec, active style master, active character registry when needed, and visible reference images loaded before generation.
Keep the production-light topology: the visible child session writes sNN-vN-brief.md, sNN-vN-prompt.md, sNN-vN-image.png, slides.md updates, and a compact handoff directly. Agent/SubAgent and packet/return files are opt-in for test/audit, repair, exception, or explicit isolated-slide work.
Keep append-only retry behavior: incomplete outputs such as s08-v2 remain historical evidence and retries use the next unused version.
Keep the write-before version scan: one successful slide/version should have one brief/prompt/image target set, and existing files must not be overwritten without explicit authorization.

10 KiB Raw Blame History