Self-Development Workflow (historical)

Status: Replaced. The original src/workflows/self-development/ engine was deleted on 2026-05-05 in epic #2402. It never wired up to consume observability data — it was an issue-implementer pipeline, not the reinforcement loop the name suggested.

What replaced it

The reinforcement-loop intent — “consume outcome / fitness / weather data, propose improvements” — is now served by the improvement_review MCP tool (built in PR 2 of epic #2402). It reads:

  • OutcomeStore (typed outcomes per call)
  • weather_report (per-CLI per-category success rates)
  • fitness-audit JSON
  • verify_audit_chain integrity summary

Threshold-gated signals fire candidate GitHub issues for human / consensus_vote review. Never auto-merges.

See improvement_review MCP tool docs and the system-review skill.

What replaced the manual issue-implementation flow

The “process open issues” workflow that the engine was supposed to automate is now the manual dogfooding-issues skill. Same conceptual flow (analyze → research → vote → implement) but executed by an agent + human checkpoint, not by an unwired engine.

Why this happened

The original engine (~7,700 LOC) was authored before the observability primitives (OutcomeStore, weather_report, LinUCB, fitness-audit) existed. By the time those landed, the engine had drifted out of the active path — no package.json / CI / CLI dispatch was invoking its runner. Per the deprecation-and-migration skill, six months of unwired existence + an in-place replacement = clean removal.

References

  • Epic: #2402
  • Deletion PR: filed under #2402
  • improvement_review build PR: filed under #2402
  • Manual replacement: skills/dogfooding-issues/SKILL.md