🦛 Hippo Memory Insights — 2026-06-11

[github-actions[bot]]

Summary

• Memories in store: 535 (482 episodic, 53 semantic)
• New lessons learned from commits today: 0 (no fix/revert/bug commits in the last 7 days)
• Consolidation: 79 episodic entries merged into 29 new semantic memories; 18 new embeddings created
• Low-quality memories (strength < 0.5): 244 — worth a pruning pass
• Highest-confidence memory: Stale .lock.yml files cause CI friction (strength=1.00, retrieved 32x)

---

Top Memories Surfaced

1. Recurring incident (strength=1.00, 32x): Stale workflow .lock.yml files cause churn and CI friction; run make recompile after markdown workflow edits.
2. MCP safety (strength=1.00, 58x): Guard list_code_scanning_alerts calls in workflow prompts — always include state: open and severity: critical,high to avoid oversized MCP responses.
3. Core identity (strength=1.00, 58x): gh-aw is a GitHub CLI extension (gh aw) that compiles markdown workflows into GitHub Actions — not the GitHub CLI itself.
4. Build discipline (strength=1.00, 58x): make build + make recompile after every code or workflow change; never skip recompile.
5. Two-checkpoint validation (strength=1.00, 49x): make build && make fmt after first edit (Checkpoint 1); make agent-report-progress before every PR (Checkpoint 2).
6. Recurring incident (strength=1.00): node: command not found on GPU/self-hosted paths when Node.js runtime is missing; validate runtimes and PATH early.
7. CI failure pattern (strength=1.00): Missing make fmt before PR has caused 5 CI failures in a single day — PRs that fail CI after opening are closed without merging.

---

Suggested Improvements

Error Patterns

• Stale .lock.yml files (verified, strength=1.00, 32x retrieved): The single highest-confidence recurring incident. Consider a pre-commit hook or CI check that validates lock files are current after any .md workflow edit.
• Codex auth failures (strength=0.06, fading): Auth/mode/token setup is fragile on rerun. Add a quick auth check at the start of agent jobs to fail fast before spending compute.
• node missing on GPU/self-hosted runners (strength=1.00): Add a node --version sanity check early in workflows that depend on Node.js.

Code Quality

• JavaScript refactoring: skills/javascript-refactoring/SKILL.md documents splitting large .cjs files. Any file > 300 lines with 2+ distinct domains should be split. Worth an audit pass.
• Validation refactoring: scratchpad/validation-refactoring.md has a step-by-step guide. Check pkg/workflow/compiler_yaml.go and related validators for candidates.
• Type patterns: Loose string/int mixing is a recurring bug source. Use distinct named types (e.g., JobName vs StepID) for compile-time safety.

CI Health

• Pre-PR lint gate: The two-checkpoint strategy is well-established in memory but still causing failures. Add it as a mandatory checklist item in PR templates.
• make test vs selective testing: Memory consistently recommends letting CI run make test (>5 min) rather than running locally. Reinforce in onboarding docs.
• validate-yaml job: Ensures no ANSI escapes in YAML before other CI jobs run — keep this at the top of the dependency graph in .github/workflows/ci.yml.

Quick Wins

1. Lock-file freshness check — add a make recompile reminder to the PR template or a CI job that validates .lock.yml freshness after .md edits. Directly addresses the top recurring incident.
2. Prune low-strength memories — 244 memories (46%) have strength < 0.5. Run hippo audit and bulk-delete entries below 0.1 to improve recall precision.
3. MCP alert query guard — enforce state: open + severity: critical,high in all list_code_scanning_alerts calls. Add a linter rule or document in AGENTS.md.
4. Node.js runtime validation — add node --version || (echo 'node not found' && exit 1) early in all workflows using Node.js tooling.
5. JavaScript refactoring skill visibility — the skill has only 22 retrievals; if large .cjs files exist, link the skill from AGENTS.md to increase adoption.

Longer-term Themes

• Recompile discipline: The pattern of missing make recompile after workflow edits appears in 4+ independent memories. A Git post-commit hook may be more reliable than documentation alone.
• Memory store health: 244 of 535 memories (46%) have strength < 0.5. A regular pruning workflow or lower import threshold would keep the store focused and recall more precise.
• Semantic consolidation is healthy: 79 episodic entries merged into 29 semantics this cycle. Over time, high-signal patterns will stabilize as semantics while low-signal entries decay naturally.

---

Memory Health

Metric	Value
Total memories	535
Episodic	482
Semantic	53
Low quality (strength < 0.5)	244 (46%)
Active incident memories	3
New embeddings this cycle	18
Memories shared to global store	245

Low-quality memories flagged for pruning

Run hippo audit for the full list. Examples of fading duplicates:

• mem_ef4574760dbb (strength=0.07) — duplicate of make build/fmt checkpoint reminder
• mem_6fb77abf55fb (strength=0.07) — duplicate of JavaScript fmt-cjs reminder
• mem_b3de87b76481 (strength=0.06) — Codex auth failure incident (consider archiving if not recurring)

References: 27332590144

Generated by 🦛 Daily Hippo Learn · 274.7 AIC · ⌖ 13.7 AIC · ⊞ 20.5K · ◷

• expires on Jun 14, 2026, 12:08 AM UTC-08:00

[github-actions[bot]]

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

A newer discussion is available at Discussion #39032.