🦛 Hippo Memory Insights — 2026-05-30

[github-actions[bot]]

Summary

• Memories in store: 533 (480 episodic, 53 semantic)
• New lessons learned from commits today: 1 (OTLP docs update)
• Merged episodic → semantic: 29 new patterns consolidated
• Auto-shared to global store: 243 high-value memories
• Highest-confidence memory: gh-aw vs GitHub Copilot CLI distinction (retrieved 25×, strength 1.00)

---

Top Memories Surfaced

1. [25×] gh-aw is a GitHub CLI extension (gh aw) — not the GitHub Copilot CLI. Do not conflate the two when writing workflow prompts. (tag: important-note-gh-aw-vs-github-copilot-cli)
2. [25×] Guard list_code_scanning_alerts calls: always include state: open and severity: critical,high to avoid oversized MCP responses. (tag: checkpoint-2)
3. [25×] After merging main: run make build && make recompile — rebuilds binary and recompiles all workflow lock files. (tag: merging-main-branch)
4. [22×] Run make build && make fmt immediately after the first substantial code edit. Cheap (~2s), gives immediate feedback. (tag: checkpoint-1)
5. [20×] TWO-CHECKPOINT VALIDATION STRATEGY — NO EXCEPTIONS: checkpoint 1 after first edit, checkpoint 2 before every report_progress/PR. (tag: mandatory-pre-commit-and-pre-pr-validation)
6. [20×] Unformatted code causes immediate CI failures that block all other work — make fmt is non-negotiable before committing.
7. [20×] Stale .lock.yml files are the top recurring CI incident: always run make recompile after any markdown workflow change. (score 0.950 — highest CI signal)

---

Suggested Improvements

Error Patterns

• Stale lock files are the single highest-confidence CI failure pattern (score 0.950). A pre-commit hook or a make recompile step in the developer workflow could prevent this automatically.
• Missing make fmt before commits has caused 5+ CI failures in a single day. Consider enforcing format checks in a Git pre-commit hook or adding a CI fast-fail job.
• Oversized MCP responses from list_code_scanning_alerts without filters cause agent failures. Workflow templates should include state: open + severity: critical,high guards by default.

Code Quality

• scratchpad/validation-refactoring.md documents a step-by-step guide for splitting validators. Validators exceeding 300 lines or containing 2+ unrelated domains should be split.
• skills/javascript-refactoring/SKILL.md has been surfaced repeatedly for JS splitting work in actions/setup/js/. Files there should be kept under 300 lines.
• skills/developer/SKILL.md covers the full coding conventions — agents should reference it when unclear about validation architecture or file naming.

CI Health

• Recurring pattern: stale .lock.yml files after markdown changes → add make recompile to the contribution checklist and/or as a CI gate.
• validate-yaml job in .github/workflows/ci.yml scans for ANSI escape sequences — ensure no debug output leaks into YAML/lock files.
• make test-unit takes ~3 minutes. Prefer selective testing (go test ./pkg/specific/...) during development; let CI run the full suite.

Quick Wins

1. Add pre-commit hook for make recompile — directly targets the rejig docs #1 recurring CI incident (stale lock files).
2. Add pre-commit hook for make fmt — prevents the unformatted-code CI failures that have caused 5 failures/day.
3. Audit 52 low-strength memories (strength < 0.2) — run mcpscripts hippo --args "audit" and prune stale entries to keep the store precise.
4. Add state: open + severity: critical,high to all list_code_scanning_alerts workflow uses — prevents oversized MCP payloads.
5. Document the two-checkpoint strategy more visibly in AGENTS.md — it is the most-recalled memory (25×) and agents still miss it.

Longer-Term Themes

• Validation architecture refactoring: Multiple memories point to large validators needing decomposition. A systematic pass through pkg/parser/ validators against the 300-line rule would reduce long-term debt.
• Skill lazy-loading discipline: The "BE LAZY with skills" rule surfaces repeatedly. Consider tooling or linting to warn when skill files are pre-loaded unnecessarily.
• Runtime file copying complexity (actions/setup/js/ → /tmp/gh-aw/actions): This pattern has generated confusion. Better documentation or a diagram of the copy flow would reduce agent mistakes.

---

Memory Health

View Memory Store Details

Metric	Count
Total memories	533
Episodic	480
Semantic	53
Low strength (< 0.2)	52
Low quality (audit flagged)	7
Merged this cycle	79 episodic → 29 semantic
Decayed (removed)	0
Auto-shared to global	243

Pruning candidates: 52 memories with strength < 0.2, mostly cluster around the checkpoint-2 tag group — these are older episodic variants that have been superseded by consolidated semantic memories. Running mcpscripts hippo --args "audit" will identify which to drop.

References:

• Workflow run: §26677826328

Generated by 🦛 Daily Hippo Learn · sonnet46 1.8M · ◷

• expires on Jun 2, 2026, 7:23 AM UTC

[github-actions[bot]]

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

A newer discussion is available at Discussion #36081.