🦛 Hippo Memory Insights — 2026-06-14

[github-actions[bot]]

Summary

• Memories in store: 534 (481 episodic, 53 semantic)
• New lessons learned from commits today: 0 (no fix/revert/bug commits in last 7 days)
• Merges during consolidation: 79 episodic → semantic, 29 new semantic patterns
• Highest-confidence memory: Recurring incident: stale workflow .lock.yml files cause churn and CI friction; run make recompile after markdown workflow edits (verified, 39 retrievals)

---

Top Memories Surfaced

1. gh-aw identity (retrieved 70×) — gh-aw is a GitHub CLI extension (gh aw) that compiles markdown workflows into GitHub Actions — not the GitHub Copilot CLI.
2. Guard list_code_scanning_alerts (retrieved 70×) — Always include state: open and severity: critical,high to prevent oversized MCP responses.
3. Merge-branch rebuild (retrieved 70×) — After merging main: run make build then make recompile to keep lock files in sync.
4. Checkpoint 1 (retrieved 58×) — make build && make fmt immediately after the first substantial code change catches compile and format issues early (~2 s, cheap).
5. ANSI sanitization (retrieved 56×) — pkg/workflow/compiler_yaml.go auto-strips ANSI codes from descriptions/sources; the validate-yaml CI job enforces this before other jobs run.
6. Two-checkpoint validation strategy (retrieved 55×) — Checkpoint 1 after first edit, Checkpoint 2 (make agent-report-progress with zero lint errors) before every PR. CI will fail if skipped.
7. Stale .lock.yml incident (verified, 39×) — Recurring CI friction caused by stale lock files; make recompile is the fix after any .md workflow edit.

---

Suggested Improvements

Error Patterns

• Stale .lock.yml files are the single most-confirmed recurring incident in the store. Automatically running make recompile as part of a pre-push hook (or a lint step in CI) would eliminate the class entirely.
• node: command not found on GPU/self-hosted runners — validate that Node.js runtime is available before workflow steps that require it; add a guard check in runner setup.
• Oversized MCP responses from list_code_scanning_alerts — enforce state: open + severity: critical,high filters by adding validation in prompt templates or workflow linting.

Code Quality

• JavaScript refactoring — skills/javascript-refactoring/SKILL.md documents splitting large .cjs files. Files >300 lines or containing 2+ distinct domains are candidates for splitting.
• Validation complexity — scratchpad/validation-refactoring.md has a step-by-step guide; the decision tree (>300 lines → split, 2+ domains → split) is already encoded as a memory but not yet checked by any linter.
• Type safety — Use typed slices/strong IDs (JobName, StepID, WorkflowID) to prevent mixing incompatible string types; GitHubAllowedTools pattern is a good reference.
• Unformatted code causes immediate CI failures. Consider adding a fast make fmt --check step that runs locally before git commit (pre-commit hook).

CI Health

• CI validates YAML for ANSI escape sequences (validate-yaml job in .github/workflows/ci.yml) before other jobs. Developers should run make fmt locally to avoid this gate.
• Full test suite (make test) >5 min — let CI run it; avoid local runs during development. No flakiness signals in the current memory set.
• PRs failing CI immediately are closed without merging — a wasted agent session. The two-checkpoint strategy prevents this.

Quick Wins

1. Pre-commit hook for make recompile — eliminates the top recurring CI incident (stale .lock.yml). One-line hook, low risk.
2. Pre-commit hook for make fmt — prevents "unformatted code → immediate CI failure" (5+ CI failures in a single day observed).
3. Prompt-level guard for list_code_scanning_alerts — add state: open, severity: critical,high as required fields in workflow prompt templates.
4. Lint rule: file size — implement the >300-line split threshold as an automated linter check (see scratchpad/validation-refactoring.md).
5. Node.js availability check — add a node --version || exit 1 guard step to workflows that depend on Node, preventing the node: command not found incident on self-hosted runners.

Longer-term Themes

• Lock-file drift appears multiple times across incident, error, and CI memories. A CI step that detects mismatched lock files and fails fast with a clear message would address root cause.
• Validation complexity surfaces repeatedly. A dedicated refactoring effort on large validators (>300 lines) would reduce cognitive load and improve testability.
• Skills discovery — multiple memories enforce "be lazy, don't preload skills." As the library grows, an index-based discovery mechanism would help agents choose the right skill without reading all of them.
• MCP response size management — the list_code_scanning_alerts guard is a known pattern; a generalised MCP response-size linting pass could identify other unbounded queries.

---

Memory Health

• Total: 534 memories (481 episodic, 53 semantic)
• Decayed / removed: 0 this cycle
• Low-quality flagged by audit: 7 (run hippo audit for details)
• Auto-shared to global store: 245 high-value memories promoted
• Embeddings: 534/534 current (19 new embeddings created this cycle)

Memories worth pruning

The audit flagged 7 low-quality memories. Run hippo audit to inspect and prune them. Additionally, 3 memories with strength=0.30 (decay underway) are candidates for review:

• mem_96e76c2ea9ef — testing guidelines pointer (low strength, high retrieval)
• mem_7dfbfdeeff10 — merge conflict fix note (low strength)
• mem_46a55aef27a0 — "when to use each approach" guidance fragment (low strength)

---

References: §27492421573

Generated by 🦛 Daily Hippo Learn · 249.2 AIC · ⌖ 13.6 AIC · ⊞ 21.1K · ◷

• expires on Jun 16, 2026, 11:58 PM UTC-08:00

[github-actions[bot]]

This discussion has been marked as outdated by Daily Hippo Learn.

A newer discussion is available at Discussion #39749.