Repository navigation
Description
Turns the Gap Analysis: Kiro by AWS (Discussion #1118) into buildable work. The analysis found this repo already ahead of Kiro on model routing, canary rollout, eval-gated promotion, and scheduled-automation maturity, and identified a small set of genuine gaps. Critically, one recommended phase — MCP curation (source-analysis Phase 1) — is already delivered: epic #676 wired the opt-in MCP knob into scripts/engine.sh and #816 promoted the zero-auth Context7 pilot (see docs/initiatives/mcp-powered-review.md). This initiative therefore does NOT re-plan MCP; it (1) reconciles the analysis against the current repo so no already-done gap is rebuilt, (2) ships the one clearly-recommended new build — spec-drift detection (gap G1) — on top of the existing initiative pipeline, and (3) produces a design-only spike for the large/risky multi-repo coordination gap (G3) so a human can decide whether to open a separate build epic.
The build decisions for the contested gaps (developer-local hooks G5, session-continuity G4, multi-repo driver G3) are deliberately routed to the Phase 0 decision record's recommendation rather than presumed here.
Success metrics (how we'll know it worked)
- Reconciliation completeness: every one of the 10 Kiro capability rows in the analysis carries an explicit, repo-grounded
build/already-covered/dropverdict indocs/initiatives/kiro-gap-analysis.md. - Spec-drift quality at ship: the detector produces 0 false-positive drift advisories across the frozen held-out eval set, and is designed to surface >= 1 true-positive drift detection per month once promoted to live (source analysis section 6, Phase-2 target).
- Multi-repo decision unblocked: the spike ends with a go/no-go recommendation and, if go, a scoped follow-up epic outline — no code shipped until a human opts in.
Cost cap / budget bound
- Spec-drift adds at most one cheap (triage) engine invocation per merged
initiative-labeled PR — no cron, no polling, no new scheduled automation. It is bounded byengine.sh's existing per-tier token logging andTRIAGE_TIMEOUT_SEC(default 300s). - The Phase 0 audit and the Phase 2 spike are one-shot documents with zero recurring cost.
- The spec-drift workflow is opt-in and byte-identical no-op when disabled, so it costs nothing until a maintainer turns it on.
Untracked prerequisites
- MCP curation (source-analysis Phase 1) is already delivered — epic MCP-powered review enrichment for the self-hosted Claude review engine (engine.sh) #676 wired the engine MCP knob and Add a zero-auth MCP server (Context7) to the PR-review action agent runtime #816 promoted the zero-auth Context7 pilot (docs/initiatives/mcp-powered-review.md). This initiative does NOT re-plan it; the Phase 0 record must confirm this so it is not rebuilt.
- Actually building the multi-repo coordination gap (G3) beyond the Phase 2 design spike requires a separate, human-gated build epic — it is out of scope here by deliberate blast-radius caution.
Planned from idea discussion #1118 by the BMAD Scrum Master initiative-planner. Inert until a maintainer adds initiative:auto.
Activity
don-petry commentedon Jul 14, 2026
Reconciled + activated (2026-07-13)
Verified against
main: no spec-drift detector exists — all 5 stories genuine. The epic already accounts for the only overtaken item (MCP curation, delivered via #676/#816) and explicitly does not re-plan it; Phase 0 #1143 will record that verdict. All phases are opt-in / one-shot / low-cost (spec-drift is a byte-identical no-op when disabled; the multi-repo G3 work is deliberately bounded to the #1147 design-only spike + a human go/no-go).Activated for dev-lead: #1143 (reconciliation record), #1144/#1145/#1146 (spec-drift detector → zero-FP eval → opt-in post-merge wiring), #1147 (multi-repo spike, design-only). No new build ships for G3 without a separate human-gated epic.
Adding
initiative:auto. — Claude Codedon-petry commentedon Oct 21, 2026
Closing after the 2026-10-21 backlog triage.
Why: All 5 stories closed; spec-drift.yml on main.
Evidence: sub-issues 5/5 closed; main:.github/workflows/spec-drift.yml
Reopen if this still reproduces on main.