suriyel/longtaskforagent/long-task-retrospective

Retrospective Report — Consolidate & Upload Skill Improvement Records

Invoked conditionally by long-task-st Step 12.5 after Go verdict, when retrospective records exist and feedback is authorized. This skill consolidates collected records and uploads them to the configured REST API endpoint.

Announce at start: "I'm using the long-task-retrospective skill. Preparing to report skill improvement records."

Core principle: This skill only reports — it does NOT modify skill files. Improvement records are uploaded for analysis and potential integration into future skill releases.

Checklist

1. Gate Check

Verify both conditions:

a) Authorization:

• Read feature-list.json — check retro_authorized field
• If absent or false → print "Retrospective: not authorized — skipping" → STOP

b) Service reachability:

python scripts/check_retro_auth.py feature-list.json

• Exit 0 (ready) → proceed
• Exit 1 (unavailable) → print "Retrospective: endpoint unavailable — skipping" → STOP
• Exit 2 (disabled) → print "Retrospective: no endpoint configured — skipping" → STOP

2. Read Records

Read all docs/retrospectives/*.md files (excluding reported/ subdirectory):

python scripts/check_retrospective_readiness.py

For each record, validate:

python scripts/validate_retrospective_record.py docs/retrospectives/<file>.md

• Valid records → include in report
• Invalid records → log warning, exclude from upload

3. Summary

Compile statistics from record frontmatter:

• Total records (valid only)
• By severity: critical / important / minor
• By category: skill-gap / missing-rule / false-assumption / template-defect / process-gap
• By classification: systemic / one-off

Present summary to user.

4. User Confirmation

Use AskUserQuestion:

"本项目共搜集 {N} 条 Skill 改进记录（critical: {X}, important: {Y}, minor: {Z}）。
 其中系统性问题 {S} 条，一次性问题 {O} 条。是否上报到 {endpoint}？"
 
Options: "确认上报 (Recommended)" / "跳过本次上报"

• User chooses "跳过" → print "Retrospective: user skipped upload" → STOP
• User chooses "确认上报" → proceed

5. Upload

python scripts/post_retrospective_report.py --feature-list feature-list.json

The script:

1. Compresses docs/retrospectives/*.md into retrospectives.tar.gz
2. POSTs as multipart/form-data to the configured endpoint
3. Includes metadata: project name, date, branch, record count

• Exit 0 → print "Retrospective: {N} records uploaded successfully"
• Exit 1 → print error, do NOT retry — report the failure to user

6. Cleanup

• Move uploaded records to docs/retrospectives/reported/ (audit trail)
• Git commit: retro: upload {N} skill improvement records
• Update task-progress.md with retrospective entry

Critical Rules

• Never modify skill files — this skill only collects and reports
• Gate check is non-negotiable — both authorization AND reachability must pass
• User confirmation required — never upload without explicit user consent
• Privacy first — records must not contain project source code, business data, or credentials
• One upload per ST cycle — do not upload partial batches during Worker sessions
• Failure is non-blocking — upload failure does not affect ST verdict

Integration

Called by: long-task-st (Step 12.5, after Persist, before Verdict) Requires: retro_authorized = true in feature-list.json AND endpoint reachable Reads: docs/retrospectives/*.md, feature-list.json Produces: moves records to docs/retrospectives/reported/ Scripts used: check_retro_auth.py, check_retrospective_readiness.py, validate_retrospective_record.py, post_retrospective_report.py