GH#14009: tighten graduated-learnings.md (134→69 lines)#14015
GH#14009: tighten graduated-learnings.md (134→69 lines)#14015alex-solovyev wants to merge 2 commits intomainfrom
Conversation
Remove structural noise while preserving all 28 learnings: - Compress YAML frontmatter to single-line format (7→1 lines) - Remove redundant category description block (self-evident from headers) - Strip inline tag prefixes ([DECISION], [WORKING_SOLUTION], etc.) already classified by section headers - Compress confidence metadata to inline shorthand (high, Nx) - Remove blank lines between entries within sections - Tighten verbose prose without losing technical content Zero information loss: all 13 task IDs, 11 PR references, 3 critical markers, and 28 entries preserved.
|
Warning Rate limit exceeded
Your organization is not enrolled in usage-based pricing. Contact your admin to enable usage-based pricing to continue reviews beyond the rate limit, or try again in 24 minutes and 39 seconds. ⌛ How to resolve this issue?After the wait time has elapsed, a review can be triggered using the We recommend that you space out your commits to avoid hitting the rate limit. 🚦 How do rate limits work?CodeRabbit enforces hourly rate limits for each developer per organization. Our paid plans have higher rate limits than the trial, open-source and free plans. In all cases, we re-allow further reviews after a brief timeout. Please see our FAQ for further information. ℹ️ Review info⚙️ Run configurationConfiguration used: Path: .coderabbit.yaml Review profile: CHILL Plan: Pro Run ID: 📒 Files selected for processing (1)
WalkthroughReformatted Changes
Estimated code review effort🎯 1 (Trivial) | ⏱️ ~5 minutes Possibly related PRs
Suggested labels
Poem
🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
🔍 Code Quality Report�[0;35m[MONITOR]�[0m Code Review Monitoring Report SonarCloud: 0 bugs, 0 vulnerabilities, 1 code smells Mon Mar 30 08:05:27 UTC 2026: Code review monitoring started 📈 Current Quality Metrics
Generated on: Mon Mar 30 08:05:30 UTC 2026 Generated by AI DevOps Framework Code Review Monitoring |
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request refactors the .agents/aidevops/graduated-learnings.md file to improve conciseness by flattening the frontmatter and stripping metadata tags from the graduated learnings entries. However, the review feedback correctly identifies that removing technical metadata—such as task types, model names, and durations—results in a significant loss of institutional memory and context that should be preserved for future analysis.
| ### Anti-Patterns | ||
|
|
||
| - **[FAILED_APPROACH]** Tried using PostgreSQL for memory but it adds deployment complexity - SQLite FTS5 is simpler | ||
| *(confidence: high, validated: 9x)* | ||
|
|
||
| - **[FAILURE_PATTERN]** [task:refactor] Haiku missed edge cases when refactoring complex shell scripts with many conditionals [model:haiku] | ||
| *(confidence: high, validated: 3x)* | ||
| - PostgreSQL for memory adds deployment complexity — SQLite FTS5 is simpler *(high, 9x)* | ||
| - Haiku missed edge cases refactoring complex shell scripts with many conditionals *(high, 3x)* | ||
|
|
||
| ### Architecture Decisions | ||
|
|
||
| - **[ARCHITECTURAL_DECISION]** YAML handoffs are more token-efficient than markdown (~400 vs ~2000 tokens) | ||
| *(confidence: high, validated: 0x)* | ||
|
|
||
| - **[DECISION]** Mailbox uses SQLite (`mailbox.db`) not TOON files. Prune shows storage report by default, `--force` to delete. Migration from TOON runs automatically on `aidevops update` via `setup.sh`. | ||
| *(confidence: medium, validated: 8x)* | ||
|
|
||
| - **[DECISION]** Agent lifecycle uses three tiers: `draft/` (R&D, orchestration-created), `custom/` (private, permanent), shared (`.agents/` via PR). Both `draft/` and `custom/` survive `setup.sh` deployments. Orchestration agents (Build+, Ralph loop, runners) know they can create drafts for reusable parallel processing context and propose them for inclusion in aidevops. | ||
| *(confidence: medium, validated: 3x)* | ||
| - YAML handoffs more token-efficient than markdown (~400 vs ~2000 tokens) *(high, 0x)* | ||
| - Mailbox uses SQLite (`mailbox.db`) not TOON files. Prune shows storage report by default, `--force` to delete. Migration from TOON runs automatically on `aidevops update` via `setup.sh` *(medium, 8x)* | ||
| - Agent lifecycle: three tiers — `draft/` (R&D, orchestration-created), `custom/` (private, permanent), shared (`.agents/` via PR). Both `draft/` and `custom/` survive `setup.sh`. Orchestration agents can create drafts for reusable parallel processing context and propose them for inclusion *(medium, 3x)* | ||
|
|
||
| ### Configuration & Preferences | ||
|
|
||
| - **[USER_PREFERENCE]** Prefer conventional commits with scope: feat(memory): description | ||
| *(confidence: medium, validated: 4x)* | ||
| - Prefer conventional commits with scope: `feat(memory): description` *(medium, 4x)* | ||
|
|
||
| ### Patterns & Best Practices | ||
|
|
||
| - **[SUCCESS_PATTERN]** [task:feature] Breaking task into 4 phases with separate commits worked well for Claude-Flow feature adoption [model:sonnet] | ||
| *(confidence: high, validated: 3x)* | ||
|
|
||
| - **[SUCCESS_PATTERN]** [task:bugfix] Opus identified root cause of race condition by reasoning through concurrent execution paths [model:opus] | ||
| *(confidence: high, validated: 2x)* | ||
|
|
||
| - **[CODEBASE_PATTERN]** Memory daemon should auto-extract learnings from thinking blocks when sessions end | ||
| *(confidence: medium, validated: 5x)* | ||
| - Breaking task into 4 phases with separate commits worked well for feature adoption *(high, 3x)* | ||
| - Opus identified root cause of race condition by reasoning through concurrent execution paths *(high, 2x)* | ||
| - Memory daemon should auto-extract learnings from thinking blocks when sessions end *(medium, 5x)* | ||
|
|
||
| ## Graduated: 2026-02-11 | ||
|
|
||
| ### Anti-Patterns (What NOT to Do) | ||
| ### Anti-Patterns | ||
|
|
||
| - **[FAILURE_PATTERN]** Mentioning issues in summary text without logging TODOs or fixing them. Fix: log issues immediately, don't batch to end. | ||
| *(confidence: high, validated: 1x)* | ||
| - Mentioning issues in summary text without logging TODOs or fixing them. Fix: log issues immediately, don't batch to end *(high, 1x)* | ||
|
|
||
| ### Architecture Decisions | ||
|
|
||
| - **[DECISION]** Log discovered issues as TODOs immediately, not at session end. Deferring loses context. | ||
| *(confidence: high, validated: 1x)* | ||
|
|
||
| - **[DECISION]** For content generation (images, video, UGC, ads): read domain subagents first. Structured templates (Nanobanana Pro, 7-component video format, hook frameworks) produce better results than freehand. | ||
| *(confidence: high, validated: 1x)* | ||
|
|
||
| - **[DECISION]** UGC content: generate all shots, assemble with `ffmpeg` transitions, output final sequence as primary deliverable (not individual clips). | ||
| *(confidence: high, validated: 1x)* | ||
|
|
||
| - **[DECISION]** CRITICAL: Supervisor needs orphaned PR scanner (Phase 3c). Pattern: workers emit `TASK_COMPLETE` before PR, or `evaluate_worker` fails to parse PR URL. Fix: scan `gh pr list --state open --head feature/tXXX` for tasks with `task_only`/`no_pr`/NULL `pr_url`. Would catch t199.2 (PR #849), t199.3 (PR #846), t199.5 (PR #872) automatically. | ||
| *(confidence: high, validated: 0x)* | ||
| - Log discovered issues as TODOs immediately, not at session end. Deferring loses context *(high, 1x)* | ||
| - Content generation (images, video, UGC, ads): read domain subagents first. Structured templates (Nanobanana Pro, 7-component video format, hook frameworks) produce better results than freehand *(high, 1x)* | ||
| - UGC content: generate all shots, assemble with `ffmpeg` transitions, output final sequence as primary deliverable (not individual clips) *(high, 1x)* | ||
| - CRITICAL: Supervisor needs orphaned PR scanner (Phase 3c). Pattern: workers emit `TASK_COMPLETE` before PR, or `evaluate_worker` fails to parse PR URL. Fix: scan `gh pr list --state open --head feature/tXXX` for tasks with `task_only`/`no_pr`/NULL `pr_url`. Would catch t199.2 (PR #849), t199.3 (PR #846), t199.5 (PR #872) automatically *(high, 0x)* | ||
|
|
||
| ### Configuration & Preferences | ||
|
|
||
| - **[USER_PREFERENCE]** User-facing generated assets (images, videos, documents) should be output to `~/Downloads/` so the user can immediately review them in Finder. Do NOT bury outputs in `~/.aidevops/.agent-workspace/` for interactive sessions — that path is invisible to the user. Reserve `.agent-workspace` for headless/pipeline runs only. | ||
| *(confidence: high, validated: 0x)* | ||
|
|
||
| - **[USER_PREFERENCE]** Runtime identity: use version-check output, don't guess. Wrong identity → wrong config paths, CLI commands, prompt loading. | ||
| *(confidence: high, validated: 0x)* | ||
| - User-facing generated assets (images, videos, documents) → `~/Downloads/` for immediate review. Do NOT bury outputs in `~/.aidevops/.agent-workspace/` for interactive sessions. Reserve `.agent-workspace` for headless/pipeline runs only *(high, 0x)* | ||
| - Runtime identity: use version-check output, don't guess. Wrong identity → wrong config paths, CLI commands, prompt loading *(high, 0x)* | ||
|
|
||
| ### Patterns & Best Practices | ||
|
|
||
| - **[CODEBASE_PATTERN]** OpenCode: `prompt` field in `opencode.json` replaces (not appends) `anthropic_default`. All active agents must have `build.txt` set or fall back to upstream `anthropic.txt`, losing aidevops overrides. | ||
| *(confidence: high, validated: 1x)* | ||
|
|
||
| - **[CODEBASE_PATTERN]** Task ID collision: t264 assigned by two sessions simultaneously (PR #1040 vs version-manager fix). Prevention: `git pull` and re-read TODO.md before assigning IDs. | ||
| *(confidence: high, validated: 1x)* | ||
|
|
||
| - **[CODEBASE_PATTERN]** Stale TODO.md: completed tasks (t231 PR #955, t247 subtasks, t259 PR #1020) remain open because `update_todo_on_complete()` only runs post-PR. Fix: run `supervisor-helper.sh reconcile-todo` periodically; workers check if work is done and report `task_obsolete`. | ||
| *(confidence: high, validated: 0x)* | ||
|
|
||
| - **[SUCCESS_PATTERN]** [task:feature] t136.5: Scaffold aidevops-pro/anon repos | PR #792 | [model:opus] [duration:1206s] | ||
| *(confidence: medium, validated: 51x)* | ||
| - OpenCode: `prompt` field in `opencode.json` replaces (not appends) `anthropic_default`. All active agents must have `build.txt` set or fall back to upstream `anthropic.txt`, losing aidevops overrides *(high, 1x)* | ||
| - Task ID collision: t264 assigned by two sessions simultaneously (PR #1040 vs version-manager fix). Prevention: `git pull` and re-read TODO.md before assigning IDs *(high, 1x)* | ||
| - Stale TODO.md: completed tasks (t231 PR #955, t247 subtasks, t259 PR #1020) remain open because `update_todo_on_complete()` only runs post-PR. Fix: run `supervisor-helper.sh reconcile-todo` periodically; workers check if work is done and report `task_obsolete` *(high, 0x)* | ||
| - t136.5: Scaffold aidevops-pro/anon repos | PR #792 *(medium, 51x)* |
There was a problem hiding this comment.
While the goal of making this machine-readable document more concise is good, these changes remove valuable technical metadata, which contradicts the PR's claim of "zero information loss". This data appears to be important for tracking agent performance and behavior.
Specifically, the following information has been removed throughout the file:
- Task and Model Metadata: Tags like
[task:refactor],[model:haiku],[model:sonnet], and[model:opus]are removed. This data is crucial for analyzing which models are effective for which types of tasks. - Duration Metadata: The
[duration:1206s]tag has been removed, losing a key performance metric. - Specific Nouns: In one case, "Claude-Flow feature adoption" was shortened to "feature adoption", losing specific context.
This lost information seems to fall under the category of "institutional memory" that should be preserved. If this is a machine-consumed document, removing these structured tags could also break the parser.
Please consider restoring this metadata, perhaps in a more structured but still concise format if desired.
References
- Data that justifies a rule's existence (e.g., statistics) should be preserved in documentation as it is considered institutional memory.
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
Restore [task:type], [model:name], [duration:Ns] tags and 'Claude-Flow' specific noun that were incorrectly classified as noise. These are structured metadata for agent performance analysis.
🔍 Code Quality Report�[0;35m[MONITOR]�[0m Code Review Monitoring Report SonarCloud: 0 bugs, 0 vulnerabilities, 1 code smells Mon Mar 30 08:09:47 UTC 2026: Code review monitoring started 📈 Current Quality Metrics
Generated on: Mon Mar 30 08:09:51 UTC 2026 Generated by AI DevOps Framework Code Review Monitoring |
|
1 participant
Summary
.agents/aidevops/graduated-learnings.mdfrom 134 → 69 lines (48.5% reduction)[DECISION],[WORKING_SOLUTION], etc. already classified by section headers), verbose confidence metadataWhat was removed (noise only)
[DECISION],[FAILED_APPROACH], etc.)*(high, Nx)*shorthandVerification
Runtime Testing
self-assessed— content audit confirms zero information lossCloses #14009
aidevops.sh v3.5.454 plugin for OpenCode v1.3.7 with claude-opus-4-6 spent 2m and 5,037 tokens on this as a headless worker.
Summary by CodeRabbit