learning-aggregator

# Learning Aggregator Reads accumulated `.learnings/` files across all sessions, finds patterns, and produces a ranked list of promotion candidates. This is the outer loop's **inspect** step. Without this skill, `.learnings/` is a write-only log. Patterns accumulate but nobody synthesizes them. The same gap resurfaces two weeks later because no one looked. ## When to Use - **Weekly cadence** — scheduled or manual, review accumulated learnings - **Before major tasks** — check if the task area has known patterns - **After a burst of sessions** — consolidate findings from a sprint or incident - **When self-improvement flags `promotion_ready`** — verify the flag with full context ## What It Produces A **gap report** — a ranked list of patterns that have crossed (or are approaching) the promotion threshold, with evidence and recommended actions. ## Step 1: Read All Learning Files Read these files in `.learnings/`: | File | Contains | |------|----------| | `LEARNINGS.md` | Corrections, knowledge gaps, best practices, recurring patterns | | `ERRORS.md` | Command failures, API errors, exceptions | | `FEATURE_REQUESTS.md` | Missing capabilities | Parse each entry's metadata: - `Pattern-Key` — the stable deduplication key - `Recurrence-Count` — how many times this pattern has been seen - `First-Seen` / `Last-Seen` — date range - `Priority` — low / medium / high / critical - `Status` — pending / promotion_ready / promoted / dismissed - `Area` — frontend / backend / infra / tests / docs / config - `Related Files` — which parts of the codebase are affected - `Source` — conversation / error / user_feedback / simplify-and-harden - `Tags` — free-form labels ## Step 2: Group and Aggregate Group entries by `Pattern-Key`. For each group: 1. **Sum recurrences** across all entries with the same key 2. **Count distinct tasks** — how many different sessions/tasks encountered this 3. **Compute time window** — days between First-Seen and Last-Seen 4. **Collect all related files** — union of all entries' file references 5. **Take highest priority** across entries in the group 6. **Collect evidence** — the Summary and Details from each entry For entries without a `Pattern-Key`, use conservative grouping only: - **Exact match**: Same `Area` AND at least 2 identical `Tags` - **File overlap**: Same `Related Files` path (exact path match, not substring) - **Do NOT fuzzy-match** on Summary text — false groupings are worse than ungrouped entries Flag ungrouped entries separately with a recommendation to assign a `Pattern-Key`. Ungrouped entries are common and expected — they may be one-off issues or genuinely novel problems. ## Step 3: Rank and Classify ### Promotion Threshold An entry is **promotion-ready** when: - `Recurrence-Count >= 3` across the group - Seen in `>= 2 distinct tasks` - Within a `30-day window` ### Approaching Threshold An entry is **approaching** when: - `Recurrence-Count >= 2` or - `Priority: high/critical` with any recurrence ### Classification For each promotion candidate, classify the gap type: | Gap Type | Signal | Fix Target | |----------|--------|------------| | **Knowledge gap** | Agent didn't know X | Update project instruction files (CLAUDE.md, AGENTS.md, .github/copilot-instructions.md) | | **Tool gap** | Agent improvised around missing capability | Add or update MCP tool / script | | **Skill gap** | Same behavior pattern keeps failing | Create or update a skill (use `/skill-creator`, validate with `quick_validate.py`, register `skill-check` eval) | | **Ambiguity** | Conflicting interpretations of spec/prompt | Tighten instructions or add examples | | **Reasoning failure** | Agent had the knowledge but reasoned wrong | Add explicit decision rules or constraints | ## Step 4: Produce Gap Report Output a structured report: ```markdown ## Learning Aggregator: Gap Report **Scan date:** YYYY-MM-DD **Period:** [since date] to [now] **Entries scanned:** N **Patterns found:** N **Promotion-ready:** N **Approaching threshold:** N ### Promotion-Ready Patterns #### 1. [Pattern-Key] — [Summary] - **Recurrence:** N times across M tasks - **Window:** First-Seen → Last-Seen - **Priority:** high - **Gap type:** knowledge gap - **Area:** backend - **Related files:** path/to/file.ext - **Evidence:** - [LRN-YYYYMMDD-001] Summary of first occurrence - [LRN-YYYYMMDD-002] Summary of second occurrence - [ERR-YYYYMMDD-001] Summary of related error - **Recommended action:** Add rule to project instruction files (CLAUDE.md, AGENTS.md, .github/copilot-instructions.md): "[concise prevention rule]" - **Eval candidate:** Yes — [description of what to test] #### 2. ... ### Approaching Threshold #### 1. [Pattern-Key] — [Summary] - **Recurrence:** 2 times across 1 task - **Needs:** 1 more recurrence or 1 more distinct task - ... ### Ungrouped Entries (no Pattern-Key) - [LRN-YYYYMMDD-005] "Summary" — needs pattern_key assignment - ... ### Dismissed / Stale - Entries with Last-Seen > 90 days ago and Status: pending → recommend dismissal ``` ## Step 5: Handoff The gap report feeds into: 1. **harness-updater agent** — takes promotion-ready patterns and applies them to project instruction files (CLAUDE.md, AGENTS.md, .github/copilot-instructions.md) 2. **eval-creator skill** — takes eval candidates and creates permanent test cases 3. **Human review** — for patterns classified as "reasoning failure" or "ambiguity" (these need human judgment) ## Filtering - `--since YYYY-MM-DD` — only scan entries after this date - `--min-recurrence N` — raise the promotion threshold - `--area AREA` — filter to a specific area (frontend, backend, etc.) - `--deep` — also analyze session traces (see Session Trace Analysis below) ## Session Trace Analysis The outer loop reads from two complementary sources: | Source | What it is | Cadence | Cost | |--------|-----------|---------|------| | `.learnings/` | Explicit entries written by self-improvement during sessions. Agent's own reflections: corrections, knowledge gaps, recurring patterns it noticed. | Every session (hot path) | Near-zero | | Session transcripts | Full session transcripts from GitHub Actions `agent` artifacts: prompts, tool calls, outputs, token usage. Available for all gh-aw factory workflows. | Weekly or on-demand (cold path) | Moderate — download per run | The default mode reads `.learnings/` and produces a gap report from what the agent explicitly logged. The `--deep` mode also analyzes session transcripts and merges findings from both sources. ### Why both sources matter `.learnings/` captures what the agent **noticed and chose to log** — a curated subset. Session transcripts capture **everything that happened**, including patterns the agent worked around, retried, or never recognized as failures. Examples of patterns visible in transcripts but absent from `.learnings/`: - **Retry loops**: The same tool call repeated 3+ times with small variations. The agent eventually got it right but never logged the initial failures. - **Noop patterns**: Workflows that called noop on runs that should have produced output — a signal of misconfigured triggers or overly strict noop conditions. - **Worked-around test failures**: A test failed, the agent changed approach, the new approach passed, the original failure was forgotten. - **Context handoff causes**: Which drift signals actually triggered handoffs, not just that handoffs happened. - **Token/time anomalies**: Sessions with disproportionate cost vs output — a signal of inefficiency the agent is unaware of. - **Spec drift**: Agent spending effort on out-of-scope work, visible in tool call sequences before a pivot. These patterns are high-value for the outer loop because the agent can't self-report them. Session transcripts are the only source. ### When to trigger --deep mode Trace analysis is **not** per-session. It's cadenced: - **Weekly scheduled** (recommended minimum): after a sprint or burst of sessions - **Post-incident**: when something went wrong and you want to understand why - **Pre-promotion**: before committing a pattern to project instruction files, verify it actually recurs in real sessions - **Manual invocation**: `/learning-aggregator --deep --since 7d` Running trace analysis per-session would burn tokens without producing new signal — cross-session patterns only emerge over multiple sessions. ## GitHub Actions Transcript Analysis Every factory workflow compiled with gh-aw uploads an `agent` artifact after the agent step completes. This artifact contains the full session transcript and is the primary source for `--deep` mode analysis. ### Artifact contents | File | What it contains | |------|-----------------| | `agent-stdio.log` | Full conversation: the prompt, all tool calls, tool outputs, and agent reasoning in chronological order | | `sandbox/agent/logs/` | Structured agent logs with timestamps and tool metadata | | `safeoutputs.jsonl` | Structured record of every safe-output action the agent took (issue created, comment posted, etc.) | | `agent_output.json` | The final structured output payload | | `agent_usage.json` | Token usage: prompt tokens, completion tokens, total | ### Discovering artifacts Use the GitHub CLI to list recent runs and download artifacts: ```bash # List recent runs for a specific factory workflow gh run list --workflow spec-refiner.lock.yml --limit 10 \ --json databaseId,displayTitle,conclusion,createdAt,event,headBranch,headSha # Download the agent artifact for a specific run mkdir -p /tmp/transcripts/<run-id> gh run download <run-id> --name agent --dir /tmp/transcripts/<run-id> # Or via the API gh api repos/{owner}/{repo}/actions/runs/{run-id}/artifacts ``` Artifact retention is 90 days by default (the gh-aw default). After 90 days, the artifact is deleted automatically. ### What to extract from a transcript For each `agent-stdio.log` file, parse the conversation and look for: 1. **Tool call repetition** — same tool + similar args called 3+ times in sequence → likely a retry loop. Pattern-key: `retry-loop.<tool>` 2. **Noop on actionable input** — agent called noop but the triggering event clearly warranted action → Pattern-key: `noop-misfire.<workflow>` 3. **Error patterns in tool output** — responses containing `error`, `failed`, `Traceback`, `not found` before the agent recovered → Pattern-key: `error.<category>` 4. **Approach changes mid-task** — agent abandoning a path and restarting (visible as repeated similar tool calls with different parameters after an error) → Pattern-key: `approach-switch.<domain>` 5. **Token anomalies** — `agent_usage.json` showing token count more than 2x the median for similar workflows → Pattern-key: `cost.<workflow>` 6. **Spec drift signals** — tool calls accessing files or making changes clearly outside the stated scope → Pattern-key: `drift.<workflow>` Each finding is mapped to the same taxonomy as self-improvement: - `harden.*` — security, validation, permissions - `simplify.*` — complexity, dead code, over-abstraction - `process.*` — workflow ordering, handoff logic - `spec.*` — scope adherence, plan compliance ### Privacy handling Transcripts may contain content from issue bodies, commit messages, and PR descriptions. These can include PII (names, email addresses, code snippets from private contexts). When analyzing: - Extract only the **structural patterns** (tool call sequences, error categories, retry counts) - Do not copy raw transcript content into issues or `.learnings/` entries - Do not include issue body excerpts unless they are already public on GitHub - Summarize patterns in abstract terms: "agent retried file-read 5 times before succeeding" not the actual file content ### How the two sources merge in the gap report When `--deep` runs, each pattern in the gap report gets a `sources` field: ```yaml promotion_ready: - pattern_key: "retry-loop.file-read" recurrence_count: 5 sources: - .learnings/LEARNINGS.md (2 entries) - transcript:spec-refiner/run-12345678 (3 occurrences) confidence: high # appears in both sources evidence: - "LRN-20260401-001: File read retry on large repos" - "transcript:12345678: Same grep tool called 4 times with varying patterns" - "transcript:12345679: File not found on first attempt, succeeded on second" ``` A pattern in both sources is higher confidence than one from either alone. ### Reading traces with Entire (optional) If [Entire](https://entire.io) is installed and enabled on this repo, the `--deep` flag also uses the Entire CLI for local Claude Code session transcripts: ```bash # Check availability entire --version # List recent checkpoints as JSON entire rewind --list # Read a checkpoint's full transcript entire explain --checkpoint <id> --full --no-pager ``` If `entire` is not installed, `--deep` uses only GitHub Actions artifact transcripts as described above. Entire and Actions artifact analysis are complementary: | Source | Covers | Best for | |--------|--------|----------| | GitHub Actions artifacts | All gh-aw factory workflow runs | Automated factory patterns | | Entire checkpoints | Local Claude Code sessions | Human-driven interactive patterns | ## Persistence Reads `.learnings/` from the working directory. This is the only persistence mode — the skill does not integrate with external memory backends in interactive sessions. For CI-side durable storage across workflow runs, see `learning-aggregator-ci`, which can optionally back its state with gh-aw's `repo-memory` (git-branch persistence). The resulting branch is a normal git branch and can be fetched locally if desired, but the interactive skill itself only reads local files. ### Tracker-id in gap reports Each promotion candidate in the gap report includes a `tracker` field set to the pattern-key. This tracker propagates through the full chain: harness-updater embeds it as a comment in project instruction files, eval-creator references it in eval cases. To audit the full lifecycle of a pattern, search for `tracker:[pattern-key]` across the repo and GitHub. ## What This Skill Does NOT Do - Does not modify `.learnings/` files (read-only analysis) - Does not apply promotions (that's harness-updater) - Does not create evals (that's eval-creator) - Does not fix code or run tests - Does not replace human judgment for ambiguous patterns - Does not run `--deep` trace analysis per-session — only on cadence or explicit invocation - Does not require Entire — falls back to `.learnings/`-only mode when trace source is unavailable