Skip to content

feat: /adr skill — architectural decision records with decision gate#737

Open
ljagged wants to merge 12 commits intogarrytan:mainfrom
ljagged:feat-adr
Open

feat: /adr skill — architectural decision records with decision gate#737
ljagged wants to merge 12 commits intogarrytan:mainfrom
ljagged:feat-adr

Conversation

@ljagged
Copy link
Copy Markdown

@ljagged ljagged commented Apr 1, 2026

Summary

  • New /adr skill with 5 modes (create, list, check, revisit, litmus-test) plus utilities (accept, renumber, skipped)
  • Decision gate in shared preamble detects architectural decisions in-flight and pauses before implementation — the key differentiator vs. every other ADR tool
  • {{ADR_DISCOVERY}} resolver consumed by /plan-eng-review, /review, /plan-ceo-review, and /investigate
  • ADR count in review readiness dashboard
  • ADRs stored in docs/adr/ (repo, not ~/.gstack/) with YAML frontmatter
  • Skip log in docs/adr/.skipped.jsonl (JSONL, CI-friendly)
  • Sensitivity config via repo-local docs/adr/.config (not global gstack-config)
  • Adaptive tradeoff probing: brief for irrelevant dimensions, persistent for avoided ones
  • Supersession diffs show what changed between old and new decisions

Design decisions

  • Storage in repo (docs/adr/): ADRs are project artifacts. Full visibility of drafts matters as much as finalized records.
  • Preamble gate, not PreToolUse hooks: Behavioral instructions at tier 2+, same pattern as proactive routing. Self-excludes from /adr itself to prevent recursion.
  • Conditional injection: Gate payload only present when docs/adr/ exists. Zero overhead for non-ADR projects.
  • LLM-as-parser: Bash grep for resolver counts, LLM reads files via Read tool for check/list. No npm YAML dependency.
  • Sequential numbering: Community standard. Race conditions across branches accepted; /adr renumber for post-merge cleanup.
  • Status model: User chooses proposed vs. accepted on creation. /adr accept N for explicit transition.

Test plan

  • bun run gen:skill-docs generates cleanly (adr: 869 lines, ~10049 tokens)
  • bun test passes (602 pass, 4 fail — all pre-existing in strategist)
  • Dogfooded on lifeexamined-app project — created ADRs, tested litmus-test flow
  • E2E: planted contradiction detection (periodic tier, post-merge)
  • E2E: litmus-test verdict accuracy (periodic tier, post-merge)

🤖 Generated with Claude Code

divergentflow-old and others added 12 commits March 26, 2026 13:33
…hestration (v1.1.0)

Two modes: brief (autonomous competitive intelligence via WebSearch + browse)
and session (interactive Rumelt's kernel diagnosis with framework selection).
Produces versioned strategy documents with inline citations, milestone-gated
execution plans, and change tracking. Integrates with the gstack skill network.

v1.1 improvements from dogfood: broad market scan for competitor discovery,
browse-aggressive research, coherent actions with mutual support structure,
milestone-gated execution (not calendar months), brief amendment step, Codex
second opinion option, and "what did I miss?" validation.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Both review skills now discover *-strategy-*.md files from /strategist
alongside design docs. Strategy docs provide competitive context and
guiding policy to inform scope and architecture decisions during reviews.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
CLAUDE.md: add /strategist to available skills list.
TODOS.md: add 3 items — CSO/retro artifact ingestion, autoplan
integration, and E2E eval test (all post-dogfood).

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Standardize on gstack-slug everywhere (was mixing remote-slug and
  gstack-slug, causing Mode 1 write / Mode 2 read path mismatch)
- Add SLUG derivation to bash blocks that used it without re-deriving
- Fix description: "milestone-gated execution plans" not "90-day action plans"
- Add auto-chain retry guard: if Mode 1 fails, stop instead of retrying

Found by adversarial review (Claude subagent, 15 findings).

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
feat: add /strategist skill — competitive strategy with framework orchestration (v0.11.22.0)
Resolved conflicts in VERSION, CHANGELOG.md, README.md.
Kept upstream version (0.12.11.0), upstream CHANGELOG entries,
and upstream's --single-branch --depth 1 clone command.
Preserved /strategist in skill lists from feat-strategist merge.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
…(v0.1.0)

New skill with 5 modes (create, list, check, revisit, litmus-test) plus
utilities (accept, renumber, skipped). Decision gate in shared preamble
detects architectural decisions in-flight and pauses before implementation.
ADRs stored in docs/adr/ with YAML frontmatter. Skip log in .skipped.jsonl.

Includes {{ADR_DISCOVERY}} resolver consumed by /plan-eng-review, /review,
/plan-ceo-review, and /investigate. ADR count in review readiness dashboard.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
# Conflicts:
#	README.md
#	SKILL.md
#	SKILL.md.tmpl
#	TODOS.md
#	canary/SKILL.md
#	cso/SKILL.md
#	design-shotgun/SKILL.md
#	document-release/SKILL.md
#	investigate/SKILL.md
#	retro/SKILL.md
#	scripts/resolvers/index.ts
#	scripts/resolvers/preamble.ts
#	setup-deploy/SKILL.md
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants