Skip to content

[Phase 0] Kiro capability reconciliation + gap decision record #1143

Closed
#1214
Closed
[Phase 0] Kiro capability reconciliation + gap decision record#1143
#1214
Labels
dev-leadFor dev-lead agent pickupinitiativeEpic / initiative tracking issue

Description

@github-actions
github-actions
bot
opened on Jul 10, 2026

Story

As a maintainer weighing which Kiro-inspired gaps to close,
I want a repo-grounded decision record that reconciles each capability in the Kiro gap analysis against what this repo already has and recommends build / already-covered / drop for every remaining gap,
so that we invest only in genuinely-missing, high-value capabilities and never rebuild something (like MCP curation) that already shipped.

Acceptance Criteria

  1. A new doc docs/initiatives/kiro-gap-analysis.md exists, following the initiative-doc header shape used by the existing docs under docs/initiatives/ (Status / Author / Date / Scope / Constraints).
  2. The doc reproduces the 10-row capability comparison from Discussion Gap Analysis: Kiro by AWS #1118 and, for EACH row, records a verdict of already-covered, build, or drop with a one-line justification citing the concrete repo artifact that grounds it (e.g. scripts/engine.sh, scripts/initiative-driver.sh, scripts/initiative_canary.sh, evals/).
  3. The MCP-curation gap (source-analysis Phase 1) is explicitly marked already-covered, citing epic MCP-powered review enrichment for the self-hosted Claude review engine (engine.sh) #676, Add a zero-auth MCP server (Context7) to the PR-review action agent runtime #816, and docs/initiatives/mcp-powered-review.md, so it is not re-planned.
  4. Developer-local agent hooks (gap G5) are assessed against Claude Code's actual hook support: the doc states which of Kiro's 8 trigger types Claude Code hooks already cover (via settings.json) and gives a build/drop recommendation grounded in that finding — not an assumption.
  5. The remaining gaps G1 (spec-drift), G3 (multi-repo coordination), G4 (session continuity), and G5 (dev-local hooks) each carry an explicit go/no-go recommendation and a rough size, so a human can pick what to build next.
  6. The doc does not recommend adopting any AWS-proprietary Kiro infrastructure (Bedrock routing, Powers marketplace, Autonomous Agent) as a dependency — consistent with the repo's multi-provider engine strategy and the analysis's vendor-lock-in risk.

Tasks / Subtasks

Dev Notes

  • This is a documentation/decision story — no scripts or workflows change. Match the header shape and tone of the existing docs/initiatives/*.md (see mcp-powered-review.md and idea-to-initiative-pipeline.md) so it slots into the initiative-doc set.
  • Ground every verdict in a real artifact. Key anchors: model routing = scripts/engine.sh (per-tier chains + fallback); canary rollout = scripts/initiative_canary.sh; eval gates = evals/ + evals/deep-review/holdout; scheduled automation = the .github/workflows/*.yml set (feature-ideation, initiative-driver, actions-fleet-monitor); MCP = scripts/engine.sh (REVIEW_MCP_CONFIG) + docs/initiatives/mcp-powered-review.md.
  • MCP curation is DONE — do not open it as a gap. Add a zero-auth MCP server (Context7) to the PR-review action agent runtime #816 already ran a controlled A/B and found ~0 quality lift on this repo's Bash/YAML PRs; the doc should record that as the reason MCP is already-covered (and where Context7's marginal value actually concentrates), not re-litigate it.
  • For G5, the check is whether Claude Code's settings.json hooks (UserPromptSubmit, PreToolUse, PostToolUse, Stop, etc.) already cover Kiro's developer-local trigger types. State coverage from evidence; do NOT decide to build a .claude/hooks/ convention here — that decision is the doc's output, and if it lands as build it becomes a separate future story.
  • Testing standard: this is prose, so there is no bats surface; the readiness bar is that every row's verdict cites a path that exists in the repo.

Project Structure Notes

Adds one file under docs/initiatives/, consistent with how prior initiative decision records are stored. No code paths touched.

References

  • docs/initiatives/mcp-powered-review.md
  • docs/initiatives/idea-to-initiative-pipeline.md
  • scripts/engine.sh
  • scripts/initiative-driver.sh
  • scripts/initiative_canary.sh
  • evals/deep-review/holdout
  • AGENTS.md

Likely target surface

  • docs/initiatives/

Story prepared by the BMAD Scrum Master (Bob) for epic #1142. Status: ready-for-dev.

Activity

  1. github-actions
    added
    initiativeEpic / initiative tracking issue
    on Jul 10, 2026
  2. github-actions
    added a parent issue on Jul 10, 2026
  3. github-actions
    marked this as blocking 2 issues on Jul 10, 2026
  4. don-petry
    added
    dev-leadFor dev-lead agent pickup
    on Jul 14, 2026

  5. don-petry commented on Jul 14, 2026

    @don-petry
    don-petry
    on Jul 14, 2026
    Collaborator

    Dev-Lead Implementation Plan

    Issue: #1143 — [Phase 0] Kiro capability reconciliation + gap decision record

    Scope

    Add one decision-record doc, docs/initiatives/kiro-gap-analysis.md, that reconciles every capability row and gap from Discussion #1118 against concrete repo artifacts and records a build / already-covered / drop verdict for each, so only genuinely-missing capabilities get built.

    Implementation Checklist

    • Create docs/initiatives/kiro-gap-analysis.md with the initiative-doc header shape (Status / Author / Date / Scope / Constraints), matching mcp-powered-review.md.
    • Reproduce the capability comparison table from Discussion Gap Analysis: Kiro by AWS #1118, adding a Verdict column (already-covered / build / drop) + one-line grounded justification citing a real path per row.
    • Mark MCP curation (G2 / source Phase 1) already-covered, citing epic MCP-powered review enrichment for the self-hosted Claude review engine (engine.sh) #676, Add a zero-auth MCP server (Context7) to the PR-review action agent runtime #816, and docs/initiatives/mcp-powered-review.md (the ~0-quality-lift A/B), so it is not re-planned.
    • Assess G5 dev-local hooks against Claude Code's actual settings.json hook events vs. Kiro's live trigger types (fetched from kiro.dev/docs/hooks — 7 agent-lifecycle + 3 file = 10). State coverage from evidence, give build/drop.
    • Record explicit go/no-go + rough size (S/M/L) for G1 (spec-drift), G3 (multi-repo), G4 (session continuity), G5 (dev-local hooks).
    • Add the vendor-lock-in guardrail note: do not adopt AWS-proprietary Kiro infra (Bedrock routing, Powers marketplace, Autonomous Agent) as a dependency.

    Tests to Write

    • None — this is a pure documentation/decision story with no bats surface (Phase 2 TDD exception). Readiness bar: every verdict cites a repo path that exists.

    Test Command

    bash .dev-lead/scripts/dev-lead-lint.sh (shellcheck + agent frontmatter; no scripts change, but run to confirm no regressions)

  6. don-petry commented on Jul 14, 2026

    @don-petry
    don-petry
    on Jul 14, 2026
    Collaborator

    Dev-Lead: Implementation Complete

    Plan Execution

    • Created docs/initiatives/kiro-gap-analysis.md with the initiative-doc header shape (Status / Author / Date / Scope / Constraints), matching mcp-powered-review.md.
    • Reproduced the Gap Analysis: Kiro by AWS #1118 capability comparison (all 11 rows) with a Verdict column (already-covered / build / drop) + a one-line justification citing a real repo path per row.
    • Marked MCP curation already-covered, citing epics MCP-powered review enrichment for the self-hosted Claude review engine (engine.sh) #676 / Add a zero-auth MCP server (Context7) to the PR-review action agent runtime #816 and docs/initiatives/mcp-powered-review.md (the ~0-quality-lift A/B on Bash/YAML PRs) — recorded so it is not re-planned.
    • Assessed G5 against Claude Code's actual settings.json hook events vs. Kiro's live trigger types (fetched from kiro.dev/docs/hooks: 7 agent-lifecycle + 3 file). Built a coverage table and grounded the DROP recommendation in it — not an assumption.
    • Recorded explicit go/no-go + rough size for G1 (GO, M), G3 (GO spike / NO-GO full-build, L), G4 (NO-GO/drop, S–M), G5 (DROP capability, S docs-only).
    • Added the vendor-lock-in guardrail: Bedrock routing, Powers marketplace, and Autonomous Agent are each marked drop as dependencies.

    Test Results

    [lint] shellcheck on 80 file(s)...
[lint] validating 4 agent profile(s)...
OK: agents/agentic-workflows.md
OK: agents/compliance-auditor.md
OK: agents/feature-ideator.md
OK: agents/pr-reviewer.md
[lint] validating .github/CODEOWNERS...
OK: .github/CODEOWNERS
[lint] all checks passed   (rc=0)

Readiness bar (every verdict cites a path that exists): all 21 cited paths verified present.

    Files Changed

    • docs/initiatives/kiro-gap-analysis.md: new decision record reconciling every Kiro capability row and gap against grounded repo artifacts with build/already-covered/drop verdicts.

    Notes

    • Phase 2 (TDD) exception: pure documentation story — no bats/test surface. Per the story's testing standard, the readiness bar is that every verdict cites a repo path that exists; verified for all 21 paths.
    • Faithful-reproduction note: the source table in Gap Analysis: Kiro by AWS #1118 §2 actually has 11 rows (the story rounds it to "10-row"). All 11 are reproduced rather than dropping one; the doc's summary notes "10 of 11 rows are already-covered."
    • Evidence correction: the analysis referenced a canary-rollout.sh that does not exist in-repo; row 7 is grounded instead in the real ring-rollout artifacts (scripts/cut-release.sh + channel tags in docs/release/versioning.md, plus the live canaries). Kiro's hooks were "8 trigger types" in the analysis but the live docs enumerate 10 (7 agent-lifecycle + 3 file) — the doc uses the live set.
    • No scripts or workflows changed; git diff is the single new file.
  7. don-petry
    added a commit that references this issue on Jul 14, 2026

    feat: implement issue #1143 — [Phase 0] Kiro capability reconciliatio…

    a713226
  8. don-petry
    closed this as completedin #1214on Jul 14, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    dev-leadFor dev-lead agent pickupinitiativeEpic / initiative tracking issue

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions