Enhance skill-research-process for CLI tool skills

Fact-Check

Claims checked: 5 VERIFIED: 5 | REFUTED: 0 | INCONCLUSIVE: 0

All claims confirmed against file content (filesystem check — no web verification needed, these are structural claims about repo files):

1. CONFIRMED — No local path input handling. Evidence: SKILL.md:4 argument-hint is tool-or-library-name; agent-prompts.md:34-39 lists only web-based sources; no Read/Glob tools used anywhere.
2. CONFIRMED — No named reference file templates. Evidence: agent-prompts.md:63-69 prescribes only category/index.md structure with no named standard files.
3. CONFIRMED — No assets/ directory production. Evidence: absent from SKILL.md and agent-prompts.md entirely; gaps-analysis.md does not mention it either.
4. CONFIRMED — No sync/release-tracking script. Evidence: absent from SKILL.md and agent-prompts.md; gaps-analysis.md:29 explicitly documents this gap.
5. CONFIRMED — Output uses references/{category}/index.md (subdirectory). Evidence: SKILL.md:22,75,112-113 and agent-prompts.md:63-68,135 consistently use this pattern.

RT-ICA

Goal: Fix skill-research-process so it produces CLI tool skills matching the structural quality of the uv skill — with local path input, named reference file templates, assets/ output, a sync script, and flat reference file layout.

Decision: APPROVED

Conditions:

1. Local directory path input handling | AVAILABLE | Fix: update agent-prompts.md with local-path detection variant
2. Named CLI reference file templates | AVAILABLE | Fix: new references/cli-tool-reference-templates.md defining standard files (cli_reference.md, configuration.md, migration-guide.md, quick-reference.md, troubleshooting.md); update categorization agent prompt to reference it
3. Assets/ directory production step | AVAILABLE | Fix: add to Stage 3 in agent-prompts.md
4. Sync/release-tracking script production | AVAILABLE | Fix: add Stage 4 Post-Integration to agent-prompts.md delegating to @python3-development:python-cli-architect
5. Flat reference file layout | AVAILABLE | Fix: update Stage 2, Stage 3, Quality Gate 2 across SKILL.md and agent-prompts.md
6. Embed vs separate skill | RESOLVED | False dilemma — references/ directory handles this. New content goes in references/cli-tool-reference-templates.md and updates to agent-prompts.md. No new skill, no embedding into SKILL.md body.

Groomed (2026-02-25)

Priority

9/10 — Blocks CLI tool skill creation at production quality. Without these fixes, /skill-research-process produces structurally different output than the reference uv skill, missing 5 critical components (local paths, named reference files, assets, sync scripts, flat layout). Affects any skill targeting tools like ty, ripgrep, fd, where comprehensive documentation is essential.

Impact

• Blocks: New CLI tool skills cannot match existing quality standards; users get incomplete documentation
• Bottleneck: Reference uv skill demonstrates the target structure, but skill-research-process cannot produce it

Benefits

• Enables local documentation ingestion (e.g., .claude/worktrees/ty/docs/) as input
• Produces standard CLI reference files (cli_reference.md, configuration.md, migration-guide.md, quick-reference.md, troubleshooting.md) instead of arbitrary categories
• Generates assets/ directories for templates, examples, and reusable resources
• Produces sync/release-tracking scripts for keeping skills current (GitHub Releases API integration)
• Aligns output to flat references/ structure (agentskills spec standard), avoiding subdirectory navigation overhead

Expected Behavior

When invoking /skill-research-process .claude/worktrees/ty/docs/:

1. Skill detects argument is local path (starts with ./, /, ~, or contains /)
2. Reads directory structure and file index from local path
3. Passes local documentation as primary source to categorization agent
4. Categorization agent maps findings to 5 named reference file types (CLI tools)
5. Research agents populate references/{category-slug}.md (flat, not subdirectories)
6. Stage 3 creates assets/ directory with example configs, Dockerfiles, or templates
7. Stage 4 creates scripts/sync_{tool-name}_releases.py for GitHub Releases tracking
8. Output skill matches uv structural quality and completeness

Acceptance Criteria

1. Local input detection — Stage 0 reads local directory paths; categorization agent prompt template includes path-detection variant
2. Named reference file templates — New references/cli-tool-reference-templates.md defines standard file types; categorization agent maps to them
3. Assets directory production — Stage 3 Integration step creates assets/ with reusable templates or examples
4. Sync script production — Stage 4 Post-Integration creates sync script delegating to @python3-development:python-cli-architect
5. Flat reference layout — All updates to SKILL.md, agent-prompts.md, Stage 2 and Stage 3 use references/{slug}.md instead of references/{category}/index.md
6. Verified against uv skill structure — New flow produces output structurally identical to plugins/python3-development/skills/uv/ (verified by file diff)

Resources

• Skill: /skill-research-process (skill being enhanced)
• Skill: /plugin-creator:skill-creator (skill structure, frontmatter format, and validation requirements — load before modifying any SKILL.md)
• Skill: /plugin-creator:agentskills (references/ structure standard)
• Skill: /python3-development:python3-development (orchestration guide for sync script delegation)
• Agent: @python3-development:python-cli-architect (sync script production)
• Reference: .claude/skills/skill-research-process/references/agent-prompts.md (will be updated)
• Reference: .claude/skills/skill-research-process/references/cli-tool-reference-templates.md (NEW)
• Reference: plugins/python3-development/skills/uv/ (quality benchmark)
• Prior work: .claude/plan/skill-research-process-assessment.md (gap analysis)
• Prior work: .claude/skills/skill-research-process/references/gaps-analysis.md

Dependencies

• Blocks: CLI tool skill creation for ty, ripgrep, fd etc. until complete
• Depends on: None

Effort

High — 5 interdependent fixes across SKILL.md, agent-prompts.md, new reference file, new stage, and verification against uv skill structure.

Reproducibility

The gaps are reproducible and currently present:

1. Local directory input — Attempt /skill-research-process .claude/worktrees/ty/docs/ and observe that the skill treats the path as a tool name rather than reading the local directory
2. Reference file structure — Generated output uses references/{category}/index.md structure instead of named files like references/cli_reference.md
3. Missing assets/ — Generated skill will lack an assets/ directory for templates and examples
4. Missing sync script — Generated skill will lack scripts/sync_{tool-name}_releases.py for release tracking
5. Flat layout absence — References use subdirectory structure, not flat named files

Evidence files: .claude/skills/skill-research-process/SKILL.md (line 4: argument-hint is tool-or-library-name), agent-prompts.md (lines 34-39: only web-based sources; lines 63-69: only category/index.md structure)

Story

As a developer, I want The skill-research-process skill has sound research orchestration but lacks o... so that backlog items are tracked in GitHub.

Description

The skill-research-process skill has sound research orchestration but lacks output specification for producing complete CLI tool skills. Three gaps identified via assessment against the uv skill: (1) No local directory input — passing a path like .claude/worktrees/ty/docs/ is treated as a tool name, triggering web searches instead of reading local docs. (2) No CLI reference file templates — no structural anchor ensuring standard reference types (cli_reference.md, configuration.md, migration-guid

Acceptance Criteria

• Work matches description
• Plan or implementation complete

Context

• Source: Session observation
• Priority: P1
• Added: 2026-02-25
• Research questions: None