| name | adr-contribute |
| description | Harvests architectural decisions and writes lean Architecture Decision Records to docs/adr/. Works in two modes — from a dev-strategy plan (reads decision-log.yaml files) or from a raw codebase (mines git history, comments, and docs). Use when a plan is done or when you want to capture past decisions from any codebase. |
| skills-depend-on | ["contribution-system","design-principles"] |
ADR Contribute Skill
Wrap up a dev-strategy plan by surfacing the decisions that deserve to be remembered, getting human confirmation on which ones to record, and writing lean ADRs.
What This Skill Does
Harvests architectural decisions from wherever they live — structured decision-log.yaml files in a dev-strategy plan, or git history, comments, and docs in a raw codebase — then applies a three-gate filter to identify which decisions are ADR-worthy:
- Hard to reverse — changing your mind later would be costly
- Surprising without context — a future reader would wonder "why did they do it this way?"
- Real trade-off — genuine alternatives existed; one was chosen for specific reasons
The skill proposes candidates but the human selects which ones get written. When wrapping a plan, it also captures completion status — which phases were implemented, which were intentionally skipped and why.
Works with: Plans created by dev-strategy skill in .plans/plan-[FEATURE-NAME]/, or any codebase directly (no plan required)
Context & Philosophy
When Used on a Plan, This Skill Signals Completion
Invoking this skill on a dev-strategy plan means the plan is done. Not necessarily fully implemented — a deliberate early stop is a valid outcome. The skill handles partial completion gracefully: it captures what was built, what was skipped, and why.
When used on a raw codebase (no plan), there is no completion signal — the skill simply captures past decisions.
Offer ADRs Sparingly
The three-gate filter is the only guard against ADR inflation. A decision that fails any gate should not become an ADR, even if it felt significant during implementation. The value of an ADR library degrades fast when it fills with obvious choices.
See the filter applied in detail in references/tactics.md.
Human-in-the-Loop Selection
The skill proposes; the human decides. This matters because the three-gate filter is heuristic, not mechanical — the human has context about what will genuinely surprise future readers that no automated analysis can match.
Execution Overview
To wrap a plan with durable ADRs, three outcomes must be achieved:
Outcome 1 (HARVEST): Candidate list is ready for human review
- Context detected and appropriate harvest strategy applied (plan-based or codebase)
- Each decision has been assessed against the three gates
- A ranked candidate list has been presented to the user with one-line justifications
Outcome 2 (SELECTION): Human has confirmed what to write
- User has confirmed which phases were completed vs. intentionally skipped
- User has selected which ADR candidates to write (numbers, "all", or "none")
- Skipped-phase rationale captured (if any phases were left out)
Outcome 3 (DOCUMENTATION): Plan is wrapped and committed
- Selected ADRs written to
docs/adr/ with correct sequential numbering
- Wrap-up contribution folder created with
wrap-summary.md
- ADRs under
docs/adr/ committed to git; the wrap-up contribution folder under .plans/ stays uncommitted (planning scratch)
Key Outputs
ADR files (one per selected candidate):
- Location:
docs/adr/NNNN-slug.md
- Format: lean — title + 1–3 sentence summary; optional sections only when they add genuine value
- Numbering: scan existing
docs/adr/ and increment from highest
For the ADR format, see assets/adr-template.md.
Wrap-up contribution folder (always created, even when no ADRs are written):
- Location:
.plans/plan-[FEATURE-NAME]/contributions/NNN-plan-wrap-adr-contribute/
- Contains:
wrap-summary.md — phase completion status + index of ADRs written
How to Use This Skill
For tactics for each outcome, see references/tactics.md.
Core Principles
For design philosophy (YAGNI, Sufficient Documentation, Last Responsible Moment), see design-principles skill.
Prerequisites
Before using this skill, read contribution-system skill to understand:
- Folder naming convention (NNN-phase-X-[type]-[specialty]-[agent])
- Unified decision-log YAML schema
- Directory structure (contributions/ folder required)
- Progressive disclosure reading strategy
Context detection: The skill selects a harvest strategy based on the invocation prompt:
- Plan name or path provided → plan-based harvest (structured decision-log.yaml files)
- No plan specified → codebase harvest (git history, comments, docs) — default
See references/tactics.md for the detection logic and links to each strategy.
