🦛 Hippo Memory Insights — 2026-05-07

[github-actions[bot]]

Summary

• Memories in store: 526 (499 active, 52 new semantic memories created via merging)
• New lessons learned from commits today: 0 new (1 duplicate skipped from recent commit: degrade MCP guard policy message from warning to info)
• Consolidation: 125 episodic memories merged, 25 cross-layer duplicates removed, 1 high-value memory auto-promoted to global store
• Highest-confidence memory: mem_def8db416b6e (score=1.007, verified) — stale .lock.yml files cause CI churn; run make recompile after markdown edits

---

Top Memories Surfaced

1. mem_def8db416b6e ⭐ score=1.007 (verified) — Recurring incident: stale workflow .lock.yml files cause churn and CI friction; always run make recompile after editing markdown workflow files.
2. mem_a320987c466b score=0.700 — validate-yaml job in .github/workflows/ci.yml scans all YAML files for ANSI escape sequences before other jobs run.
3. mem_78ceafb94e44 score=0.594 — Missing make fmt before commit has caused 5 CI failures in a single day — run make build && make fmt at every checkpoint.
4. mem_535cbbf98fc7 score=0.473 — Workflow Health Monitoring runbook at .github/aw/runbooks/workflow-health.md covers missing-tool errors, auth failures, MCP config issues, and safe-output problems.
5. mem_e9965d42829a score=0.672 — Pre-commit make fmt check is non-negotiable: CI will fail immediately, and PRs that fail CI on opening are closed without merging.
6. mem_2d9a17bc9d87 score=0.252 — Guard list_code_scanning_alerts calls: always include state: open and severity: critical,high to avoid oversized MCP responses.
7. mem_dca15357bb18 strength=1.00 — gh-aw (gh aw) is not the GitHub Copilot CLI — it's a separate extension that compiles markdown workflows into GitHub Actions.

---

Suggested Improvements

Error Patterns

• ANSI escape sequences in YAML: The validate-yaml CI job is in place to catch this, but the pattern keeps appearing. Teams editing workflow YAML through text editors or scripts that inject color codes should double-check output before committing. Consider adding a pre-commit hook or make target to catch this locally.
• Stale .lock.yml files (highest confidence, verified pattern): After every change to a markdown workflow file, make recompile must be run. This is a recurring source of CI friction. A pre-commit hook checking for stale lock files would eliminate this entirely.
• Oversized MCP responses: list_code_scanning_alerts without filters returns too much data. All workflow prompts using this tool should enforce state: open + severity: critical,high.

Code Quality

• Formatting discipline: make fmt is frequently skipped, causing immediate CI failures. It takes <1s and should become muscle memory. Consider adding it as an auto-format on save in editor configs (.editorconfig, VS Code settings).
• File splitting: The memory store surfaces a decision tree — files over 300 lines or containing 2+ distinct domains should be split. Worth auditing pkg/ and internal/ for files exceeding this threshold.
• Type safety patterns: Use typed string aliases (e.g., JobName, StepID) to prevent mixing incompatible identifiers — this pattern is reinforced in memories about type best-practices.
• 7 low-quality memories flagged by the audit (hippo audit) — worth a pruning pass to keep the store clean.

CI Health

• Formatting is the rejig docs #1 CI failure cause: 5 failures in a day from missing make fmt. Enforce at the PR level with a required check that only runs make fmt && git diff --exit-code.
• Full test suite is slow (>5 min): Developers should run make test-unit (~3s) locally and let CI handle make test. This knowledge is well-represented in the store but worth surfacing in onboarding docs.
• Checkpoint validation workflow: The two-checkpoint pattern (build+fmt after first edit; fmt+validate before every PR/commit) is well-documented in memories — ensure AGENTS.md reflects this as a hard requirement.

Quick Wins

1. Add make recompile as a git pre-commit hook for files matching *.md in workflow directories — eliminates the top verified incident pattern.
2. Add make fmt to CI as a fast first check (or enforce via pre-commit) — eliminates the rejig docs #1 source of CI failures.
3. Run hippo audit to inspect and prune the 7 flagged low-quality memories — keeps the store signal-to-noise ratio high.
4. Guard all list_code_scanning_alerts calls in workflow prompts with state: open + severity: critical,high filter — prevents workflow timeouts from oversized responses.
5. Update onboarding docs to clarify gh-aw vs Copilot CLI distinction — multiple memories reinforce this confusion is common.

Longer-term Themes

• Skills underutilization: Multiple memories note that skills in skills/ provide specialized knowledge but should only be loaded lazily. The recurring pattern suggests workflows sometimes load skills preemptively, increasing context cost. A skills-usage audit across workflow prompts could identify unnecessary eager-loading.
• MCP configuration fragility: The Workflow Health runbook covers MCP auth/config issues repeatedly. This suggests the MCP setup is a recurring pain point worth investing in better validation tooling or clearer setup documentation.
• Episodic-to-semantic conversion: 125 episodic memories merged this cycle into 52 semantic patterns — healthy consolidation. The store is maturing well and the high average strength indicates lessons are being reinforced.

---

Memory Health

View Memory Stats

• Total: 526 memories
• Active (post-consolidation): 499
• Decayed and removed: 0 this cycle
• Episodic merged: 125 → 52 new semantic memories
• Cross-layer duplicates removed: 25
• Low-quality memories flagged: 7 (run mcpscripts hippo --args "audit" for details)
• Auto-promoted to global store: 1 high-value memory
• Embeddings: 526/526 (fully up to date after this run)
• Memory age: Mostly fresh (<7d), steady operation, mostly episodic, high avg strength, diverse topics

---

References: §25481717270

Generated by Daily Hippo Learn · ● 5.2M · ◷

• expires on May 10, 2026, 7:22 AM UTC

[github-actions[bot]]

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

A newer discussion is available at Discussion #30956.