🦛 Hippo Memory Insights — 2026-04-23

[github-actions[bot]]

Summary

• Memories in store: 490 (440 episodic, 50 semantic)
• New lessons learned from commits today: 0 (1 duplicate skipped)
• Consolidation: 118 episodic merged, 50 new semantic memories created, 24 cross-layer duplicates removed
• Highest-confidence memory: stale .lock.yml files cause CI friction (score=1.007, retrieved 3×)

---

Top Memories Surfaced

1. mem_47a200206f0d [verified] — Recurring incident: stale workflow .lock.yml files cause churn and CI friction; run make recompile after markdown workflow edits. (retrieved 3×)
2. mem_6f57d3433efa [verified] — Recurring incident: Codex auth failures can break agent jobs; verify auth/mode/token setup before reruns. (retrieved 2×)
3. mem_ed00ab594cfa [verified] — Recurring incident: node: command not found on GPU/self-hosted paths when node runtime/tooling is missing; validate runtimes and PATH early. (retrieved 3×)
4. mem_bf872c93f014 [observed] — CI WILL FAIL if make fmt / pre-commit validation is skipped — this is automatic and non-negotiable. (retrieved 2×)
5. mem_afd5d7c7fbef [observed] — The validate-yaml job in .github/workflows/ci.yml scans all YAML files for ANSI escape sequences before other jobs run.
6. mem_004adbf3befe [observed] — Use distinct Go types (e.g. JobName, StepID) to prevent silent type-confusion bugs when mixing string identifiers.
7. mem_fc26e9553e72 [observed] — Workflow Health Monitoring runbook covers missing-tool errors, auth failures, MCP config issues, and safe-input/output problems.

---

Suggested Improvements

Error Patterns

• Stale .lock.yml files are the single highest-confidence recurring incident. Add a CI lint step or pre-commit hook that detects out-of-date lock files and fails early, rather than discovering it mid-run.
• node: command not found has been hit 3 times on self-hosted/GPU runners. Standardise on a copilot-setup-steps.yml that always installs Node 22 and verifies node --version before any workflow step that needs it.
• Codex auth failures (2× retrieved): document the required env vars and token scopes in a checklist that runs during agent job setup, rather than after a failure.

Code Quality

• Type safety (JobName, StepID, WorkflowID): the memory store has multiple entries about needing distinct Go types to prevent identifier mix-ups. Audit pkg/workflow/ for places that still use raw string for workflow/job/step identifiers.
• JavaScript CJS refactoring: skills/javascript-refactoring/SKILL.md is flagged — ensure .cjs split is complete and make lint-cjs passes cleanly.
• 7 low-quality memories flagged by hippo audit — run hippo audit --fix to prune junk and improve recall precision.

CI Health

• validate-yaml job catches ANSI escape sequences; the compiler (pkg/workflow/compiler_yaml.go) strips them automatically — confirm both sides are in sync to avoid false positives.
• make test is >5 min and marked "avoid locally" — consider caching test binaries or parallelising to keep the full-suite feedback loop under 3 min in CI.
• Pre-commit validation (make fmt) is explicitly marked non-negotiable. Consider adding a GitHub Actions push lint job that blocks PRs early if formatting is missed.

Quick Wins

1. Add make recompile to the PR checklist or a CI check to detect stale .lock.yml — this alone would eliminate the rejig docs #1 recurring CI incident.
2. Run hippo audit --fix to remove 7 low-quality memories and improve recall quality.
3. Pin Node 22 in copilot-setup-steps.yml for self-hosted runners to eliminate node: command not found failures.
4. Add a Codex auth pre-flight check — a 2-line script that validates the token before the agent job starts.
5. Embed all memories (hippo embed) — currently only 4/490 memories are embedded, which limits semantic search quality.

Longer-term Themes

• Stale generated files (.lock.yml, compiled JS): multiple memories point to the same root cause — generated artifacts that drift out of sync. A broader solution would be a CI job that re-runs all code-generation steps and fails if the working tree is dirty.
• Runner environment fragility: node not found, auth failures, and missing PATH entries all point to inconsistent runner environments. A standardised copilot-setup-steps.yml with smoke tests would address this class of problems systematically.
• Memory embedding coverage is 4/490 (< 1%) — this severely degrades semantic recall. Scheduling hippo embed as part of the daily runner would unlock much better memory retrieval quality.

---

Memory Health

Store Statistics

Metric	Value
Total memories	490
Episodic	440
Semantic	50
Pinned	0
At risk (< 0.2 strength)	0
Open conflicts	0
Avg strength	1.00
Embedded	4 / 490
Low-quality flagged	7
Last sleep	2026-04-23

Action items for memory health:

• Run hippo audit --fix to remove 7 flagged low-quality entries
• Run hippo embed to increase embedding coverage from 4/490 → full store (required for better semantic recall)
• No stale or decayed memories — store is healthy

---

References: §24821356676

Generated by Daily Hippo Learn · ● 349.8K · ◷

• expires on Apr 26, 2026, 6:58 AM UTC