🦛 Hippo Memory Insights — 2026-06-13

[github-actions[bot]]

Summary

• Memories in store: 534 (481 episodic, 53 semantic)
• New lessons learned from commits today: 0 (no fix/revert/bug commits in the last 7 days)
• Consolidation results: 1 memory decayed, 79 episodic entries merged into 29 new semantic patterns, 29 duplicates deduped, 245 high-value memories shared to global store
• Highest-confidence memory: mem_4afaa0965161 — "Recurring incident: stale workflow .lock.yml files cause churn and CI friction; run make recompile after markdown workflow edits" (score=0.896, verified, retrieved 37×)

Top Memories Surfaced

1. mem_cab727bbe2e5 (retrieved 66×, score=1.00) — gh-aw is a GitHub CLI extension (gh aw) that compiles markdown workflows into GitHub Actions — not the GitHub Copilot CLI. This identity confusion is the most-retrieved memory.
2. mem_0324334270d0 (retrieved 66×) — Guard list_code_scanning_alerts calls: always include state: open and severity: critical,high to avoid oversized MCP responses.
3. mem_3c0154ff3f2d (retrieved 66×) — After merging main: make build && make recompile to rebuild binary and recompile all workflow lock files.
4. mem_4afaa0965161 (retrieved 37×, verified incident) — Stale .lock.yml files cause recurring CI churn. Always run make recompile after editing markdown workflows.
5. mem_2704ae961669 (retrieved 55×) — make build && make fmt is Checkpoint 1 — run immediately after the first significant code change to catch compile errors early (~2s).
6. mem_512f4cbb0103 (retrieved 53×) — Compiler sanitization: pkg/workflow/compiler_yaml.go automatically strips ANSI codes from descriptions, sources, and comments — validate-yaml CI job enforces this.
7. mem_ad0159793af4 (retrieved 53×) — Use distinct named types for strings/ints (e.g., JobName vs StepID) to prevent type-confusion bugs at compile time.

Suggested Improvements

Error Patterns

• Stale .lock.yml files are the single most impactful recurring CI failure source (verified incident, score=0.896). Every markdown workflow edit must be followed by make recompile. Consider adding a CI pre-check that compares .lock.yml timestamps against source .md files.
• Unformatted code causes immediate CI failures that block all other work. The two-checkpoint validation pattern (make build && make fmt at CP1, make agent-report-progress at CP2) should be enforced. The memory store has observed 5+ CI failures in a single day from this alone.
• list_code_scanning_alerts without filters: Missing state: open + severity: critical,high guards causes agent sessions to receive oversized MCP payloads. Audit existing workflow prompts for this pattern.

Code Quality

• JavaScript refactoring: skills/javascript-refactoring/SKILL.md documents splitting large .js files into .cjs modules. Files >300 lines or spanning 2+ distinct domains should be split — the decision tree is in the skill.
• Validation complexity: scratchpad/validation-refactoring.md contains a step-by-step guide for refactoring complex validation logic. Error messages spanning multiple concern areas are a known pain point (mem_358aed1553d2).
• Type safety: Use newtype patterns (type JobName string) to catch misuse at compile time rather than relying on naming conventions alone.
• New code: Prefer typed fields for compile-time safety (flagged in frontmatter-configuration-types memories, 26 entries).

CI Health

• Lock file drift is the dominant CI health issue (see Error Patterns above).
• validate-yaml runs first in CI and scans all YAML for ANSI escape sequences — any workflow that emits raw ANSI will fail before tests even run.
• make test (>5 min) should never be run locally during development — use make test-unit (~3s) for fast iteration and let CI own the full suite.
• Two-checkpoint validation is non-negotiable: PRs that fail CI immediately after opening are closed without merging (this policy appears 4+ times in high-retrieval memories).

Quick Wins

1. Run hippo audit — 7 low-quality memories were flagged during consolidation. Prune them to keep the store clean.
2. Audit workflow prompts for unguarded list_code_scanning_alerts calls — add state: open + severity: critical,high to avoid oversized MCP responses in agent sessions.
3. Add a CI check for stale .lock.yml files — compare source .md modification times against their compiled .lock.yml counterparts. This would prevent the most-cited recurring CI incident.
4. Prune very-low-strength memories — 42 memories have strength < 0.1 and are likely redundant or outdated. Run hippo audit and follow up with hippo forget on confirmed stale entries.
5. Review scratchpad/validation-refactoring.md — it has been flagged as a high-relevance code quality resource (score=0.686, 49x retrieved) but lives in scratchpad rather than as a committed guide. Consider promoting it.

Longer-term Themes

• Episodic dominance: 90% of memories (481/534) are still episodic. The consolidation pass merged 79 episodes into 29 semantic patterns today, but continued sleep cycles will be needed to fully distil accumulated knowledge into durable semantic form.
• Checkpoint discipline as culture: The two-checkpoint validation pattern appears across 65+ memory entries (the most-tagged non-imported category). This suggests the team is still building the habit; consider codifying it into the developer onboarding docs or a pre-commit hook.
• Tool boundary enforcement: The acceptEdits / Claude Code tool-enforcement memories (53x retrieved) suggest recurring questions about which tools are allowed in which contexts. A short reference card in the repository README or AGENTS.md may reduce repeated lookups.
• Low-strength memory accumulation: 226 memories (42%) have strength < 0.3 — this is expected growth, but a quarterly pruning pass with hippo audit + hippo forget would keep recall quality high.

Memory Health

Metric	Value
Total memories	534
Episodic	481 (90%)
Semantic	53 (10%)
Low strength (< 0.3)	226
Very low strength (< 0.1)	42
Low-quality (audit flag)	7
Stale (> 30d, < 0.5 strength)	0
New embeddings today	17

Consolidation details from today’s sleep run

Active memories:  534
Removed (decayed): 1  (mem_b3de87b76481, strength 0.0451)
Merged episodic:   79
New semantic:      29
Duplicates deduped: 29
Auto-shared to global store: 245

Notable merges:

• 6 episodes → semantic: "When to [use approach X]"
• 5 episodes → semantic: "ALWAYS use [pattern Y]"
• 4 episodes → semantic: build/recompile workflow
• 4 episodes → semantic: before-commit validation
• 3 episodes → semantic: MCPServer usage patterns
• 3 episodes → semantic: fix/merge conflict patterns

Recommendations for next run: Run hippo audit to inspect the 7 flagged low-quality entries, then hippo forget on confirmed stale ones. Schedule a quarterly pruning pass for the 42 very-low-strength memories.

References:

• §27460460864

Generated by 🦛 Daily Hippo Learn · 228.3 AIC · ⌖ 26 AIC · ⊞ 20.5K · ◷

• expires on Jun 15, 2026, 11:40 PM UTC-08:00

[github-actions[bot]]

This discussion was automatically closed because it expired on 2026-06-16T07:40:25.892Z.

Closed by Workflow