🦛 Hippo Memory Insights — 2026-07-13

[github-actions[bot]]

Summary

• Memories in store: 342
• New lessons learned from commits today: 0
• Highest-confidence memory: mem_4afaa0965161 — stale workflow .lock.yml files repeatedly cause CI churn when make recompile is skipped after workflow markdown edits.

Top Memories Surfaced

• mem_4afaa0965161 — recurring stale .lock.yml drift after workflow markdown edits; recompile is mandatory.
• sem_ef67ccfd978b — make agent-report-progress is the fast pre-PR validation gate and prevents avoidable CI failures.
• mem_0324334270d0 — list_code_scanning_alerts prompts should always constrain state: open and severity: critical,high to avoid oversized responses.
• mem_f7c45403b6e1 — workflow prompts should use steps.sanitized.outputs.* directly; needs.activation.outputs.text/title/body is deprecated.
• mem_e02072c9ae07 — self-hosted/GPU paths can fail with node: command not found; validate runtime availability early.
• mem_0755d605eefd — validators over ~300 lines should be split; current guidance targets 100–200 lines per validator.

Suggested Improvements

Error Patterns

• Workflow changes still regress through stale lock files. Add or strengthen repository checks so markdown workflow edits are always paired with regenerated lock files (make recompile). Evidence: mem_4afaa0965161.
• Oversized code-scanning prompts remain a recurring failure mode. Standardize prompt templates to always pass state: open and severity: critical,high when using list_code_scanning_alerts. Evidence: mem_0324334270d0.
• Environment-sensitive failures around missing Node tooling keep surfacing on nonstandard runners. Add an explicit early runtime validation step or clearer diagnostics in workflows that assume Node is present. Evidence: mem_e02072c9ae07.

Code Quality

• Large validators remain a structural smell. Audit pkg/workflow/ and pkg/cli/ for validators exceeding the 100–200 line target, especially files near or beyond the 300-line hard limit, and split by concern with dedicated tests. Evidence: mem_0755d605eefd, mem_98f16a2f627c.
• Continue replacing deprecated workflow output references with steps.sanitized.outputs.* in workflow markdown to reduce warnings and keep authoring consistent with the compiler’s current model. Evidence: mem_f7c45403b6e1.
• Prefer stronger typing where stringly-typed identifiers can cross wires (JobName, StepID, tool names). The memory store repeatedly flags type-pattern guidance as a practical bug-prevention measure. Evidence: mem_1c38df66aa62, mem_ad0159793af4.

CI Health

• Pre-PR validation discipline is a major predictor of CI success. Reinforce make agent-report-progress as the standard fast gate in contributor docs and workflow authoring guidance. Evidence: sem_ef67ccfd978b, mem_bdc213856476.
• Some workflows still assume expansive prompt payloads. Add prompt-budget guardrails or lint checks for high-risk tools like list_code_scanning_alerts to prevent response overflows before they hit CI. Evidence: mem_0324334270d0.
• Self-hosted and GPU-oriented jobs should fail fast on missing runtimes rather than deep in execution. A lightweight startup check would make failures less flaky and more actionable. Evidence: mem_e02072c9ae07.

Quick Wins

1. Add a CI guard that fails when .github/workflows/*.md changes are not accompanied by regenerated .lock.yml files. This is repeatedly flagged by memory and already has an older open issue, suggesting the pain is real and ongoing. Evidence: mem_4afaa0965161.
2. Sweep workflow markdown examples and active prompts to replace deprecated needs.activation.outputs.text/title/body references with steps.sanitized.outputs.*. This is narrow, mechanical, and directly backed by compiler behavior. Evidence: mem_f7c45403b6e1.
3. Add prompt linting or shared helper text that enforces state: open and severity: critical,high for list_code_scanning_alerts. This directly targets a known oversized-response failure mode. Evidence: mem_0324334270d0.
4. Add an early Node/runtime preflight check to self-hosted-sensitive workflows that depend on Node-based tools. This would turn a recurring environmental failure into a quick actionable error. Evidence: mem_e02072c9ae07.
5. Refactor one validator hotspot in pkg/workflow/ or pkg/cli/ that exceeds the validation-complexity guidance, proving the split pattern and reducing maintenance risk. Evidence: mem_0755d605eefd, mem_98f16a2f627c.

Longer-term Themes

• Workflow authoring reliability is still too dependent on human memory: stale lock files, deprecated expressions, and prompt-shape mistakes recur often enough to justify stronger automated enforcement.
• The repository would benefit from more structural decomposition in validation-heavy code, especially where large files mix multiple concerns and accumulate subtle error-handling logic.
• CI friction often comes from preventable input-shape or environment assumptions rather than deep product bugs; more preflight validation and lint-time enforcement should continue paying off.

Memory Health

• Store health is generally strong: 342 total memories, average strength 0.81, 342/342 embedded.
• 58 memories are at risk (<0.2 strength), 43 are stale, and the latest sleep reported 4 low-quality memories worth auditing/pruning.
• No new git-learned lessons were added today, so the most actionable output came from consolidation and recurring historical patterns.

Detail

• Confidence mix: 2 verified, 247 observed, 50 inferred, 43 stale.
• Consolidation today merged 13 episodic memories into 5 semantic memories and deduped 5 duplicates.
• No open memory conflicts were reported.

Next actions: prioritize automation around stale lock drift, deprecated workflow expression cleanup, and prompt guardrails; only create new issues where work is not already tracked.

Generated by 🦛 Daily Hippo Learn · 25.9 AIC · ⌖ 8.27 AIC · ⊞ 5.4K · ◷

• expires on Jul 15, 2026, 11:29 PM UTC-08:00