22 KiB

Raw Blame History

Character Consistency Test

Status

Task id: 2026-06-24-character-consistency
Worker profile: video-slide-production
Running mode: prompt and generate
Status: R7 production-light route accepted; session-13 direct-child/no-Agent package for S-29/S-30/S-31 is the formal production default
Child session reuse test: passed for session-04 continuation (S-33 then S-06 in the same clean child thread)
Deprecated child session id: 019ef77d-552b-7343-abe4-88921b35d7ea (do not reuse)
Session-02 child-session id: 019ef7ca-413a-7662-9633-7fb126b7ef35
Session-02 resolved version: s21-v2 (runtime fallback; no prompt/image)
Session-03 resolved version: s21-v3
Thread-level child test: session-03
Session-03 child thread id: 019ef7d2-1a7f-7a90-8c04-0098239dfe0f
Session-03 slide-agent thread id: 019ef7d6-3ff6-7213-b68c-fd646e39f1d2
Session-03 continuation recorded Agent S-14: 019ef803-95d7-7411-8d3c-c7f2f68a32e7
Session-03 duplicate Agent S-14 observed by parent: 019ef803-328a-7e40-a0b0-0333c6b5fc6c
Session-03 continuation Agent S-25: 019ef803-e757-7041-8f4a-fc9dfce23bbd
Session-04 child thread id: 019ef876-c5b2-7202-b204-774cb5fb95fd
Session-04 resolved versions: s33-v2, s06-v2
Session-04 Agent S-33: 019ef87a-3aa1-75a2-a97c-de8f57e234e1
Session-04 Agent S-06: 019ef886-8797-7243-9427-d45065fcfc47
Session-06 child thread id: 019ef907-16b7-7a43-9924-227892aa62d3
Session-06 accepted versions: s07-v2, s08-v3, s09-v2
Session-06 Agent S-07: 019ef90a-422b-7333-9021-f19b8dd1c0b7
Session-06 incomplete Agent S-08 attempt: 019ef91a-9be1-7642-8f67-f9df7bb055f0 (s08-v2 transport failure; prompt only)
Session-06 retry Agent S-08: 019ef927-d480-7252-9650-8b0670f877d7
Session-06 Agent S-09: 019ef92d-fa13-7053-aec9-fa8abb75f8d2
Session-07 child thread id: 019ef943-52af-7893-b758-199873c94b7a
Session-07 accepted versions: s11-v2, s12-v2, s13-v2
Session-07 Agent S-11: 019ef948-aa2e-7422-b5eb-f2385958f92d
Session-07 Agent S-12: 019ef94e-beef-72e2-ab88-05cf00739e7e
Session-07 Agent S-13: 019ef955-16cb-7532-ad1f-2d5447ad90f1
Session-08 child thread id: 019ef969-66e7-7c52-a437-045371668d41
Session-08 generated versions: s15-v2, s16-v2, s17-v2, s18-v2
Session-08 visual result: s16-v2 not accepted; repair deferred to R6
Session-09 child thread id: 019ef9bc-fe96-75b3-b902-5d5ef95feca3
Session-09 parent callback target: 019ef9b5-cc39-7a42-91b1-f9b0852dc95e
Session-09 generated version: s16-v3
Session-09 result: callback-only parent wakeup observed; completion gate pass; user accepted s16-v3-image.png
Session-10 child thread id: 019ef9d0-4914-7ea0-8935-bfd9bfc1aad4
Session-10 parent callback target: 019ef9b5-cc39-7a42-91b1-f9b0852dc95e
Session-10 scope: S-19, S-20, S-22, S-23
Session-10 result: user paused after S-20 worker boundary violation; s19-v2 accepted; s20-v2, s22-v2, and s23-v2 are incomplete evidence
Session-11 child thread id: 019ef9f8-8c34-77a3-82af-9a25beb2c723
Session-11 parent callback target: 019ef9b5-cc39-7a42-91b1-f9b0852dc95e
Session-11 scope: S-20, S-22, S-23, S-24
Session-11 result: completion gate pass; generated s20-v3, s22-v3, s23-v3, s24-v2; user accepted all four images
Session-12 metric: user observed 25m48s / 134K tokens for 3 slides with Agent still used.
Session-13 metric: user observed 10m32s / 114K tokens for 3 slides with direct child session and no Agent/SubAgent.
Artifact policy: append-only next slide version; overwrite disabled by default

R7 Production-Light Active Scope

Line: production-light
Worker profile: video-slide-production
Running mode: prompt and generate
Review mode: none
Scope: S-26/S-27/S-28
Expected versions from current slide folders: s26-v2, s27-v2, s28-v2
Child session: session-12 clean visible Codex child thread 019efbb4-b366-7de3-8e02-01183b3d0b13
Parent callback target: 019efb9f-ece2-78d2-ba6a-4ca1dcfe3244
Write policy: append-only; overwrite allowed: no
Production rule: generated images are production-pass outputs by default; no per-image user review gate
Parent consumption rule: read compact child handoff after callback; inspect prompt / return / image details only on blocked, failed, warning, artifact collision, duplicate worker, or explicit user audit request
Status: complete; callback received; completion gate pass; outputs recorded as production-pass by task rule; user observed 25m48s / 134K tokens

Direct-Child Production-Light Test Scope

Line: production-light
Topology: main session -> child session directly; no Agent / SubAgent
Running mode: prompt and generate
Review mode: none
Scope: S-29/S-30/S-31
Expected versions from current slide folders: s29-v2, s30-v2, s31-v2
Child session: session-13 clean visible Codex child thread 019efbd8-f75d-7512-9952-aa3528fde581
Parent callback target: 019efb9f-ece2-78d2-ba6a-4ca1dcfe3244
Write policy: append-only; overwrite allowed: no
Test purpose: measure real time and context/token use when the child session directly writes brief, prompt, image, slides table, and compact handoff without Agent, packet, return receipt, review, or final verification pass.
Status: complete; callback received; outputs recorded as production-pass by task rule; user accepted all three images and observed 10m32s / 114K tokens. This direct-child topology is now the formal production default.

Formal Production Default

Project phase source: project.md records Workflow phase: production-light.
Default child session title: 子会话：<short task/scope>.
Default topology: main session -> named child session directly.
Default child scope: up to 3 slides unless user explicitly overrides.
Default artifacts: sNN-vN-brief.md, sNN-vN-prompt.md, sNN-vN-image.png, slides.md status update, and compact child handoff.
Default omissions: no review file, no Agent/SubAgent, no packet/return receipt, no routine final verification pass.
Retained guard: scan target slide folder before writing, choose next unused version, and keep overwrite disabled.

User Instructions

Latest third-round test instruction:

Step 1: start a clean child session and send the S-33 production request, with no review.
Step 2: continue the same child session and send the S-06 production request, with no review.
Main-session checks focus on child-session startup, internal Agent / SubAgent operation, and whether the prompt contains character and style references.
User checks final image acceptability.

Earlier reset instruction after S-05:

Run the character-consistency workflow as a process and identity-consistency test.

Earlier s05 work exposed two workflow problems:

1. The child session did not start the slide-agent itself.
2. `s05-v1-image.png` was overwritten even though a broad-run candidate already existed.

The reset rule for the next run is:

do not reuse child session 019ef77d-552b-7343-abe4-88921b35d7ea;
start a new child session;
require the new child session to start the single-slide S-21 agent itself;
use a new slide artifact version resolved from existing slides/s21/s21-v*-* files;
do not overwrite existing slide artifacts;
explicitly disable review for this S-21 test so the workflow can verify that no review file is created when review is disabled.

Slide Scope

Completed previous tests:

s05
s21
s14
s25

Completed session-04 third-round test:

s33
s06
s07
s08
s09
s11
s12
s13

Completed R7 production-light trials:

S-26/S-27/S-28 as session-12, prompt and generate, review: none, Agent-backed production-light.
S-29/S-30/S-31 as session-13, prompt and generate, review: none, direct child/no-Agent.
No active authorized scope; wait for next production decision.

Child Session Split

session-01: historical previous attempt; do not reuse
session-02: s21 fallback evidence only; no prompt/image
session-03: s21 accepted -> s14/s25 accepted; duplicate S-14 worker defect recorded
session-04: clean child thread; s33-v2 then s06-v2 accepted; mechanism pass for this scope
session-06: clean child thread; s07-v2 accepted, then s08-v3 and s09-v2 accepted; s08-v2 preserved as incomplete transport-failure evidence
session-07: clean child thread; completed and user accepted S-11/S-12/S-13 as the first direct 3-slide package test
session-08: clean child thread `019ef969-66e7-7c52-a437-045371668d41`; S-15/S-16/S-17/S-18 as first direct 4-slide package test; flow gate pass, S-16 visual repair deferred
session-09: clean child thread `019ef9bc-fe96-75b3-b902-5d5ef95feca3`; R6 Phase 1 callback-only S-16 repair generated `s16-v3`; completion gate pass, user accepted image
session-10: clean child thread `019ef9d0-4914-7ea0-8935-bfd9bfc1aad4`; R6 Phase 2 first 4-shot package for S-19/S-20/S-22/S-23; paused by user after S-20 worker boundary violation; s19-v2 accepted, remaining v2 items incomplete evidence
session-11: clean child thread `019ef9f8-8c34-77a3-82af-9a25beb2c723`; R6 Phase 2 retry package for S-20/S-22/S-23/S-24; callback received, completion gate pass, user accepted all four images
session-12: clean child thread `019efbb4-b366-7de3-8e02-01183b3d0b13`; R7 Agent-backed production-light package for S-26/S-27/S-28; user accepted all three images
session-13: clean child thread `019efbd8-f75d-7512-9952-aa3528fde581`; R7 direct-child/no-Agent production-light package for S-29/S-30/S-31; user accepted all three images; formal production default

Current session-03 continuation generated s14-v2 and s25-v2 after user accepted s21-v3-image.png; user then accepted both images and closed the test. The mechanism result is not a clean pass because two S-14 workers were created for one slide/version.

session-04 was created after the carrier rule was corrected to use a clean user-visible child thread, not a forked/inherited-context thread. It received S-33 first, then S-06 as a continuation in the same child thread.

Session-04 mechanism result:

child session was visible to user and recorded as 019ef876-c5b2-7202-b204-774cb5fb95fd;
child session started one internal one-shot worker for each slide: Agent S-33 and Agent S-06;
each worker handled exactly one slide and resolved the next unused version before writing;
no s33-v2-review.md or s06-v2-review.md exists;
prompts record gpt-image-2 Advisor / host-native mode plus style and character reference inputs;
user accepted both generated images.

Continuation rule now validated for this two-slide scope:

use the same clean child thread only inside the user-approved task scope;
child thread must start one separate one-shot slide worker per slide;
do not reuse or continue a previous slide worker;
each worker handles exactly one slide;
each slide resolves the next unused version by scanning existing files;
do not overwrite existing broad-run or accepted images.

Session id record:

session-01 child session agent id: 019ef77d-552b-7343-abe4-88921b35d7ea (deprecated / do not reuse)
session-02 child session agent id: 019ef7ca-413a-7662-9633-7fb126b7ef35
session-03 child thread id: 019ef7d2-1a7f-7a90-8c04-0098239dfe0f
session-03 Agent S-21 thread id: 019ef7d6-3ff6-7213-b68c-fd646e39f1d2
session-03 recorded Agent S-14 thread id: 019ef803-95d7-7411-8d3c-c7f2f68a32e7
session-03 duplicate Agent S-14 thread id observed by parent: 019ef803-328a-7e40-a0b0-0333c6b5fc6c
session-03 Agent S-25 thread id: 019ef803-e757-7041-8f4a-fc9dfce23bbd
session-04 child thread id: 019ef876-c5b2-7202-b204-774cb5fb95fd
session-04 Agent S-33 id: 019ef87a-3aa1-75a2-a97c-de8f57e234e1
session-04 Agent S-06 id: 019ef886-8797-7243-9427-d45065fcfc47

Continuation result:

s14-v2-prompt.md, s14-v2-image.png, and s14-v2-slide-agent-return.md exist; no s14-v2-review.md exists.
s25-v2-prompt.md, s25-v2-image.png, and s25-v2-slide-agent-return.md exist; no s25-v2-review.md exists.
s14 prompt evidence records gpt-image-2 Advisor / host-native mode, visible side/front character refs, and style module B | 高位观察.
s25 prompt evidence records gpt-image-2 Advisor / host-native mode, visible side/front/body refs, and style module A | 清晨泥泞旷野 adapted for dust/detail.
User accepted both s14-v2-image.png and s25-v2-image.png.
Defect: child handoff records only Agent S-14 019ef803-95d7-7411-8d3c-c7f2f68a32e7, but parent thread inspection also found duplicate Agent S-14 019ef803-328a-7e40-a0b0-0333c6b5fc6c; the S-14 return/image source path records the duplicate thread folder. Treat step 2 as generated-with-defect, not as workflow pass.

Session-04 result:

s33-v2-brief.md, s33-v2-prompt.md, s33-v2-image.png, and session-04/returns/s33-v2-slide-agent-return.md exist; no s33-v2-review.md exists.
s06-v2-brief.md, s06-v2-prompt.md, s06-v2-image.png, and session-04/returns/s06-v2-slide-agent-return.md exist; no s06-v2-review.md exists.
s33 prompt evidence records gpt-image-2 Advisor / host-native mode, A/D-adjacent style calibration, and visible side/front/master character refs.
s06 prompt evidence records gpt-image-2 Advisor / host-native mode, style module A / Landscape-Led Road, and visible back/walking character refs.
User accepted both s33-v2-image.png and s06-v2-image.png.
Mechanism pass: this run used a clean child session and one distinct internal SubAgent per slide; no duplicate worker was observed in this scope.

Shared Context

AGENTS.md
VIDEO_WORKBENCH.md
docs/workflows/slide-task-orchestration.md
docs/workflows/slide-task-templates.md
projects/2026-06-23-在路上/project.md
projects/2026-06-23-在路上/execution-plan.md
projects/2026-06-23-在路上/visual-system/visual-system.md
projects/2026-06-23-在路上/visual-system/style/style-prompt-master.md
projects/2026-06-23-在路上/visual-system/characters/character-reference-registry.md
projects/2026-06-23-在路上/intake/阶段五-人物一致性单镜.md
projects/2026-06-23-在路上/slides/slides.md
projects/2026-06-23-在路上/slides/broad-generation-self-audit-v1.md
investigations/2026-06-24-s14-reference-capable-generation/reviews/summary-and-rule-implications.md

Visual-System Usage

Style continuity must use visual-system/style/style-prompt-master.md.
Character continuity must use visual-system/characters/character-reference-registry.md.
Reference-capable generation must use the validated host built-in reference-by-visible-context path:
- write the prompt/spec through the gpt-image-2 Skill Advisor workflow;
- load the real reference image into the conversation context before generation;
- call host image_gen with the loaded image assigned a role such as Image 1;
- record the input path, output path, and reference-path caveat in the compact handoff or task status; only Agent-backed test/audit or repair runs need a return.

Rules

Do not invoke CCPE.
Do not ask GPT V2 for final prompts.
Do not use Minimax MCP or other external review services.
Do not treat old broad-run images as accepted finals.
Do not generate visible-character identity shots from text-only prompts.
Do not use archived character or style references unless explicitly reopened.
When a slide agent is used, it handles exactly one slide.
Child sessions and any slide agents run sequentially.
In Agent-backed test/audit, repair, or exception runs, one slide/version must have exactly one owning slide-agent. A child session must not create two workers with the same packet and same output targets; prompt/image/return ownership must stay with the same worker id.
Child-session capacity for formal production-light defaults to maximum 3 slides per child session unless the task plan explicitly changes it.
Child session carrier rule: production-light uses clean named Codex child/session threads visible to the user and recorded by thread id. The child directly writes brief/prompt/image/status/handoff by default. Slide agents are reserved for test/audit, repair, explicit review, exception, or isolated single-slide work.
Production child sessions must start as clean user-visible threads. Do not use forked or inherited-context threads unless the user explicitly requests inherited context.
Main-session production wait rule: callback-only by default. The parent dispatches, records the callback target, and stops routine polling. Time-based polling schedules are fallback only when callback tooling is unavailable or the user explicitly requests polling for a named run.
Parent-facing heartbeat/status updates should not trigger extra child-thread inspection in routine production.
Production-light closeout is minimal: version scan, no overwrite, image paths/status recorded, review absent when review: none, compact handoff before callback. The fuller completion gate is for test/audit, repair, explicit review, or exception scopes.
Slide Agents, when used, are one-shot and become terminal / closed after return; do not continue or reuse them.
Main session records the new child session id so later slides can be sent to the same child session after user approval.
Resolve the slide artifact version before each slide run by scanning existing slides/sNN/sNN-v*-* files.
Do not overwrite existing slide artifacts. If any planned target file already exists, stop and report an artifact collision.
Do not backfill missing prompt or brief files for old broad-run images as if they were original generation inputs.
In production-light, sNN-vN-review.md should not be produced unless the task plan is changed.

Expected Outputs

Current production-light expected outputs:

projects/2026-06-23-在路上/slides/sNN/sNN-vN-brief.md
projects/2026-06-23-在路上/slides/sNN/sNN-vN-prompt.md
projects/2026-06-23-在路上/slides/sNN/sNN-vN-image.png
projects/2026-06-23-在路上/slides/slides.md update
projects/2026-06-23-在路上/tasks/2026-06-24-character-consistency/session-NN/child-session-handoff.md

Do not expect sNN-vN-review.md, slide-agent packet files, or slide-agent return files in production-light unless the task explicitly overrides the default.

Legacy session-04 mechanism-test task records:

projects/2026-06-23-在路上/tasks/2026-06-24-character-consistency/task-plan.md
projects/2026-06-23-在路上/tasks/2026-06-24-character-consistency/task-log.md
projects/2026-06-23-在路上/tasks/2026-06-24-character-consistency/session-04/child-session-plan.md
projects/2026-06-23-在路上/tasks/2026-06-24-character-consistency/session-04/child-session-handoff.md
projects/2026-06-23-在路上/tasks/2026-06-24-character-consistency/session-04/child-session-continuation-s06-plan.md
projects/2026-06-23-在路上/tasks/2026-06-24-character-consistency/session-04/child-session-continuation-s06-handoff.md
projects/2026-06-23-在路上/tasks/2026-06-24-character-consistency/session-04/worker-launch-ledger.md
projects/2026-06-23-在路上/tasks/2026-06-24-character-consistency/session-04/packets/s33-v2-slide-agent-packet.md
projects/2026-06-23-在路上/tasks/2026-06-24-character-consistency/session-04/packets/s06-v2-slide-agent-packet.md
projects/2026-06-23-在路上/tasks/2026-06-24-character-consistency/session-04/returns/s33-v2-slide-agent-return.md
projects/2026-06-23-在路上/tasks/2026-06-24-character-consistency/session-04/returns/s06-v2-slide-agent-return.md

Legacy session-04 slide records:

projects/2026-06-23-在路上/slides/s33/s33-v2-brief.md
projects/2026-06-23-在路上/slides/s33/s33-v2-prompt.md
projects/2026-06-23-在路上/slides/s33/s33-v2-image.png
projects/2026-06-23-在路上/slides/s06/s06-v2-brief.md
projects/2026-06-23-在路上/slides/s06/s06-v2-prompt.md
projects/2026-06-23-在路上/slides/s06/s06-v2-image.png

Historical User Review / Sampling Plan

These reporting gates are closed historical test notes. Production-light now defaults to no per-image user review; user review happens only by explicit sampling, repair request, or test/audit override.

After s21 generation, stop and report:

new child session id;
S-21 slide-agent id;
files written;
resolved output version;
whether the child session started Agent S-21 itself;
whether review was correctly suppressed;
whether the prompt used gpt-image-2 Advisor, active character reference, and the D style module;
whether the same child session should receive s14 and s25 next.

For the s14 / s25 continuation, report:

whether the existing child thread continued work;
Agent S-14 and Agent S-25 thread ids;
whether the two new agents are separate from Agent S-21;
files written for s14-v2 and s25-v2;
whether prompts used gpt-image-2 Advisor and active character/style references;
whether review files were or were not produced, matching the continuation packet rule;
whether generation completed for both slides.

This reporting gate is closed for session-03: both images were accepted by the user, review files were correctly absent, and the duplicate-worker defect was carried forward into the next mechanism test.

For the session-04 S-33 / S-06 test, report:

whether the child session was clean and user-visible rather than forked from parent context;
session-04 child thread id;
Agent S-33 and Agent S-06 ids;
whether S-06 continued the same child session without reusing Agent S-33;
files written for s33-v2 and s06-v2;
whether prompts used gpt-image-2 Advisor and active character/style references;
whether review files were or were not produced, matching the no-review packet rule;
whether generation completed and user accepted both images.

This reporting gate is closed for session-04: the user confirmed the mechanism checks passed and accepted both s33-v2-image.png and s06-v2-image.png.

Stop Conditions

Required reference image cannot be loaded into the conversation context.
gpt-image-2 mode check no longer returns a host/advisor path.
Host image generation cannot be called.
Generated image cannot be copied into the project slide folder.
Any planned output path already exists and overwrite is not explicitly authorized.
The child reports blocked, failed, or warning, or a generated result is later judged unusable and needs an append-only retry.

22 KiB Raw Blame History