[#102] feat: bootstrap core skills (/specify-prd, /bootstrap, /setup-pm)

[rucka]

PR Information

Story/Epic: #102 / #97
Type: Feature
Labels: user story

Summary

What Changed

Added 3 new Agent Skills for project bootstrap, thinned 2 how-to files to orchestration-only, and fixed inter-task confirmation behavior in /implement:

• /specify-prd — PRD creation process skill (idempotent, hypothesis-driven questioning)
• /setup-pm — PM tool configuration capability skill (composes /record-decision)
• /bootstrap — Full project setup orchestrator (composes /specify-prd, /setup-pm, /record-decision, assess-* with graceful degradation)
• /implement fix — commit-per-task strategy now pauses for developer confirmation after each task

Why This Change

P1 story enabling end-to-end project bootstrap with skills. Completes the bootstrap lifecycle: PRD creation → project categorization → standards generation → quality gate setup → PM tool configuration. How-to files thinned to orchestrators that delegate operational detail to skills.

Story Context

User Story: #102 — Bootstrap core with skills
Epic: #97 — Agent Skills in Knowledge Base
AC Coverage: All 10 acceptance criteria covered (AC-1 through AC-10)

Changes Made

Implementation Details

• T-1 /specify-prd: template analysis → hypothesis-driven info gathering → PRD creation → review. Idempotent: detects existing PRD, offers selective update.
• T-2 /setup-pm: PM tool selection via decision framework, configuration per implementation guide, ADL recording via /record-decision. Idempotent: detects existing config.
• T-3 /bootstrap: phase-level idempotency (PRD → categorization → checklist → standards → quality gates → PM). Graceful degradation when assess-* skills not installed.
• T-4 how-to-01: 222 → 96 lines (~57% reduction). Keeps phase flow, HALT gates, approval checklist.
• T-5 how-to-02: 251 → 116 lines (~54% reduction). Keeps categorization framework (Type A/B/C), phase flow, HALT gates.
• Fix /implement: Step 1.3 updated commit strategy descriptions; Step 2.8 added inter-task confirmation flow (summary → ask → verify → return).

Files Changed

Dataset (source of truth):

• Added: dataset/.skills/process/specify-prd/SKILL.md
• Added: dataset/.skills/capability/setup-pm/SKILL.md
• Added: dataset/.skills/process/bootstrap/SKILL.md
• Modified: dataset/.skills/process/implement/SKILL.md (inter-task confirmation fix)
• Modified: dataset/.pair/knowledge/how-to/01-how-to-create-PRD.md (thinned)
• Modified: dataset/.pair/knowledge/how-to/02-how-to-complete-bootstrap-checklist.md (thinned)

Installed copies (.claude/skills/):

• Added: .claude/skills/pair-process-specify-prd/SKILL.md
• Added: .claude/skills/pair-process-bootstrap/SKILL.md
• Added: .claude/skills/pair-capability-setup-pm/SKILL.md
• Modified: .claude/skills/pair-process-implement/SKILL.md
• Modified: .claude/skills/pair-capability-write-issue/SKILL.md
• Modified: .claude/skills/pair-next/SKILL.md

Bridge files:

• Modified: .pair/knowledge/how-to/01-how-to-create-PRD.md
• Modified: .pair/knowledge/how-to/02-how-to-complete-bootstrap-checklist.md
• Modified: .github/prompts/README.md

Testing

Test Results

Quality Gate: ✅ All passing (pnpm quality-gate)
├── ts:check: ✅ passing
├── test: ✅ passing
├── lint: ✅ passing
├── prettier:fix: ✅ clean
└── mdlint:fix: ✅ clean

Testing Strategy

• SKILL.md validation: All 3 new skills have valid YAML frontmatter, complete algorithm with check→skip→act→verify pattern
• Idempotency: Each skill documents detection of existing state and skip behavior
• Graceful degradation: /bootstrap documents fallback paths when assess-* skills not installed
• Standalone readability: Both thinned how-to files readable end-to-end without skills
• Composition interfaces: /bootstrap ↔ /specify-prd and /bootstrap ↔ /setup-pm interfaces documented and consistent

Quality Assurance

Code Quality Checklist

• Skills follow established SKILL.md structure (frontmatter, composed skills, algorithm, output, HALT, graceful degradation, notes)
• How-to files retain orchestration flow and HALT gates
• No duplication between how-to and skill operational detail
• All cross-references valid (relative paths to adoption files, guidelines, assets)
• Description fields use single-line quoted strings (mdlint compliance)

Review Areas

• Composition correctness: Verify /bootstrap Phase 0 correctly composes /specify-prd, Phase 4 correctly composes /setup-pm
• Idempotency logic: Each phase's check→skip pattern correctly detects completed work
• Graceful degradation paths: assess-*, /record-decision, /setup-pm fallbacks are clear and safe
• How-to thinning completeness: No operational detail retained that should live in skill
• Inter-task confirmation: /implement Step 2.8 correctly gates on developer response

Reviewer Guide

Review Focus Areas

1. Skill structure: Each SKILL.md follows the standard pattern — YAML frontmatter, composed skills table, stepped algorithm, output format, HALT conditions, graceful degradation, notes
2. How-to ↔ skill boundary: How-to retains strategic flow (phase order, HALT gates, approval checklist). Skill has operational detail (question templates, verification steps, file operations)
3. Composition interfaces: /bootstrap documents what it passes to /specify-prd and /setup-pm. The composed skills document what they expect from callers.

Key Test Scenarios

1. Fresh project: /bootstrap detects template PRD → composes /specify-prd → categorizes → generates standards → configures PM
2. Partial bootstrap: Re-invoke /bootstrap on project with PRD + categorization already done → skips Phase 0-1, resumes at Phase 2
3. Standalone /specify-prd: Invoke directly without /bootstrap → full interactive PRD creation flow
4. Standalone /setup-pm: Invoke directly → detects existing config or guides new setup

Dependencies & Related Work

Prerequisites

• Skill infrastructure + navigator (/next) #98 (skill infrastructure — Done ✅)
• Implement task with skills (/implement, /verify-quality, /record-decision) #99 (/implement + /record-decision — Done ✅)

Enables

• Assessment skills (8 assess-* with resolution cascade) #104 (assessment skills composed by /bootstrap)

Follow-up

• P1 skills Strategic planning process skills (/plan-initiatives through /plan-stories) #103, Assessment skills (8 assess-* with resolution cascade) #104, Remaining capability skills + /verify-adoption + /next update + prompt deprecation #105 depend on this infrastructure

Closes #102

[rucka]

Code Review — #116

Reviewer: AI (pair-process-review)
Date: 2026-02-13
Story: #102 — Bootstrap core with skills
Type: Feature

Review Summary

Overall Assessment

• Approved — Ready to merge

Key Changes

• 3 new SKILL.md files: /specify-prd (process), /bootstrap (process), /setup-pm (capability)
• 2 how-to files thinned to orchestration-only (01, 02)
• /implement fix: inter-task confirmation added to commit-per-task strategy
• Bridge files updated (pair-next, write-issue, GitHub prompts)

Business Value

Completes P1 bootstrap lifecycle with skills. End-to-end project setup now possible via /bootstrap composing /specify-prd and /setup-pm.

Code Review Checklist

Functionality

• Requirements Met — all 10 AC covered (AC-1 through AC-10)
• Business Logic — idempotency (check→skip→act→verify) correct in all 3 skills
• Integration — composition interfaces documented and consistent across skills
• Error Handling — HALT conditions and graceful degradation well-defined

Code Quality

• Readability — clear phase structure, consistent formatting
• Maintainability — skills follow established pattern from #99/#100/#101
• Naming — YAML frontmatter, section headers, output format consistent
• Complexity — /bootstrap is the most complex skill but well-organized with independent phases

Technical Standards

• Style Guide — single-line YAML descriptions, standard SKILL.md structure
• Architecture — skill composition pattern validated at scale (6 composed skills in /bootstrap)
• Dependencies — no new code dependencies; all cross-references valid

Quality Gates

Quality Gate: ✅ All passing (pnpm quality-gate)
├── ts:check: ✅
├── test: ✅
├── lint: ✅
├── prettier:fix: ✅ clean
└── mdlint:fix: ✅ clean

Consistency Checks

• Dataset ↔ Installed copies — all differences are expected name qualifications (short → pair-process-/pair-capability- prefix)
• Cross-references — all relative markdown links resolve to existing files
• Squash policy — single commit on branch ✅
• Commit format — [#102] feat: follows convention ✅

Positive Feedback

• Hypothesis-driven questioning in /specify-prd is a great UX pattern — reduces developer effort
• Quality gate setup in /bootstrap Phase 3.2 (executable from day one) is forward-thinking
• Inter-task confirmation in /implement addresses a real developer experience issue
• How-to thinning preserves standalone readability — non-skill assistants can still follow the flow

Minor Issues 💡

1. how-to-02 line count: 115 lines vs target 90±15 (max 105). Over by 10 lines. Consider trimming the categorization table or condensing References section.

2. Trailing slash inconsistency in installed /bootstrap copy: directory link targets drop trailing slashes (e.g., adoption/decision-log vs dataset's adoption/decision-log/). No functional impact but minor formatting inconsistency.

AC Coverage Verification

AC	Description	Status
AC-1	/specify-prd creates PRD	✅ Phase 1-4 algorithm
AC-2	/specify-prd idempotent	✅ Phase 0 detection
AC-3	/bootstrap orchestrates	✅ Phase 0-4
AC-4	/bootstrap idempotent	✅ Phase-level file detection
AC-5	/setup-pm configures	✅ Steps 2-4
AC-6	/setup-pm + /record-decision	✅ Step 5
AC-7	/setup-pm idempotent	✅ Step 1 detection
AC-8	/bootstrap + /record-decision	✅ Phase 1, 2, 3.2
AC-9	Graceful degradation	✅ Phase 2 Step 2.2
AC-10	Thinned how-to readable	✅ Both standalone-readable

Decision

APPROVED — No critical or major issues. 2 minor items noted for consideration. All AC met, quality gates pass, patterns consistent with prior stories.