🦛 Hippo Memory Insights — 2026-06-16

[github-actions[bot]]

Summary

• Memories in store: 535 (482 episodic, 53 semantic)
• New lessons learned from commits today: 1 (improve executive objective report #39490)
• Highest-confidence memory: mem_4afaa0965161 — stale .lock.yml files cause CI friction (verified incident, retrieved 43×)
• Consolidation: 79 episodic entries merged into 29 semantic patterns; 29 duplicates deduped; 246 memories promoted to global store

---

Top Memories Surfaced

1. mem_4afaa0965161 (score=0.887, verified incident, 43×) — Stale .lock.yml files cause recurring CI churn; always run make recompile after markdown workflow edits.
2. mem_0324334270d0 (78×) — Guard list_code_scanning_alerts in workflow prompts: always include state: open and severity: critical,high to avoid oversized MCP responses.
3. mem_cab727bbe2e5 (78×) — Core identity: gh-aw compiles markdown workflows into GitHub Actions — not the GitHub Actions runner itself.
4. mem_3c0154ff3f2d (78×) — Build + lock-file discipline: make build then make recompile keeps workflow lock files in sync.
5. mem_e02072c9ae07 (incident) — node: command not found on GPU/self-hosted paths; validate runtimes and PATH early.
6. mem_2704ae961669 (64×) — Run make build && make fmt after the first significant code edit to catch compile errors early (~2 s).
7. mem_a4ef0df02dbd (62×) — Claude engine: acceptEdits mode — --allowed-tools is the effective boundary; workflow tools: and network: apply as constraints.

---

Suggested Improvements

Error Patterns

• Stale lock files are the single most-cited recurring incident (mem_4afaa0965161). Agents forget to run make recompile after editing .github/aw/**/*.md. Consider adding a pre-commit hook or CI pre-check that diffs compiled lock files and fails fast if they are out of date.
• node: command not found on non-standard runners (mem_e02072c9ae07). Workflow templates that use runtimes.node should validate PATH and runtime availability in a setup step before reaching tool-heavy agent steps.
• Oversized MCP responses from list_code_scanning_alerts (mem_0324334270d0). This guard rule is heavily recalled (78×) but still needs to be enforced in new workflow prompts — consider a linter rule in pkg/linters.

Code Quality

• JavaScript refactoring debt (mem_54c981bfc72e, score=0.719): guide for splitting large .cjs files lives in skills/javascript-refactoring/SKILL.md. Files >300 lines with 2+ distinct domains are flagged for splitting.
• Validation complexity (mem_98f16a2f627c): scratchpad/validation-refactoring.md describes a refactoring guide — worth confirming work is complete or converting remaining items to issues.
• Type patterns (mem_ad0159793af4, 62×): strong Go types (e.g., JobName, StepID) prevent class-of-bug mixing; new code should adopt typed fields for compile-time safety.
• OTLP / outcome reporting (mem_48a6fa8403bf, git-learned 2026-05-21; mem_41ef3ebcdaab, 2026-06-16): two recent commits touched outcome reporting and OTLP data quality. The memory is tagged error — worth reviewing whether the quality validator now covers all edge cases.

CI Health

• Lock-file churn: the top CI pattern (score=0.887, verified incident). A dedicated CI job that runs make recompile --check (dry-run diff) would catch this automatically before merge.
• Unformatted code (mem_31b3deb42511, 43×): make fmt failures have caused ≥5 CI failures in a single day. The two-checkpoint strategy is well-established but still being missed.
• Full test suite overhead (make test >5 min): well-understood — let CI handle it, use make test-unit locally. No action needed.

Quick Wins

1. Add make recompile --check CI step — catches stale lock files before merge; estimated <30 min to implement, eliminates the rejig docs #1 recurring incident.
2. Add pkg/linters rule for list_code_scanning_alerts — enforce state: open + severity: critical,high filter in workflow prompts; prevents oversized MCP responses.
3. Add PATH/runtime validation step to node-using workflow templates — a single early node --version || exit 1 guard eliminates the node: command not found incident class.
4. Prune low-strength memories — hippo audit flagged 7 low-quality entries; running hippo prune --below 0.1 could clean up the noisiest of the 224 low-strength entries and improve recall precision.
5. Close out scratchpad/validation-refactoring.md — if refactoring is done, remove the file; if not, convert remaining items to tracked issues.

Longer-term Themes

• Commit-driven learning is sparse: only 6 git-learned memories total, with low retrieval counts relative to imported MEMORY.md entries. Enriching commit messages with structured context would surface richer patterns.
• Episodic dominance (482/535): the store is still heavily episodic. The 29 new semantic patterns from today's sleep are a healthy trend; more consolidation cycles will grow the semantic layer.
• Checkpoint-2 tag inflation (65 memories): this tag dominates the low-strength cluster. A focused consolidation pass could reduce noise while preserving the core rule.

---

Memory Health

Metric	Value
Total memories	535
Episodic	482
Semantic	53
Low-strength (<0.3)	224 (42%)
Low-quality flagged (audit)	7
New embeddings created	19

Stale/low-quality candidates: 224 memories have strength <0.3, with a cluster around the checkpoint-2 tag. Running hippo audit and pruning entries with strength <0.1 would reduce noise. The sleep cycle deduped 29 redundant semantic patterns today.

Recent git-learned memories (6 total)

Date	Commit summary
2026-06-16	improve executive objective report (#39490)
2026-06-05	refresh models.json from models.dev catalog (#37071)
2026-06-02	copilot-sdk harness stdin wiring, SDK installation/resolution, custom-provider setup from /reflect, remove duplicates
2026-05-30	use getOrCreateToolMap helper and simplify isClaudeToolName (#35855)
2026-05-21	update outcome reporting & otlp data quality validator (#33706)
2026-05-20	update otlp.md with xml docs (#33468)

Top tag clusters (frequency)

Tag	Count
imported	527
checkpoint-2-before-every-report-progress-pr-creat	65
type-patterns-and-best-practices	41
frontmatter-configuration-types	26
channel-lifecycle-guidelines	22
console-message-formatting	19
expression-transformations-in-workflows	19
command-development-checklist	18

References: §27603972820

Generated by 🦛 Daily Hippo Learn · 383.3 AIC · ⌖ 13.6 AIC · ⊞ 20.1K · ◷

• expires on Jun 19, 2026, 12:30 AM UTC-08:00