Refactoring Tasks: holistic-linting

Refactoring Tasks: holistic-linting

Plugin: ./plugins/holistic-linting Design Spec: refactor-design-holistic-linting.md Baseline Score: 68/100 Target Score: 85+/100

Task 1: Split holistic-linting SKILL.md into 3 focused skills

Status: ❌ NOT STARTED Dependencies: None Priority: 1 Complexity: High Agent: plugin-creator:refactor-skill

Target: ./plugins/holistic-linting/skills/holistic-linting/SKILL.md Issue Type: SKILL_SPLIT

Description:

Split the 1221-line monolithic SKILL.md into 3 focused skills:

1. holistic-linting (router/core) — Trimmed to ~430 lines containing: purpose (lines 1-48), linter detection and running formatters (lines 247-375), bundled resources (lines 993-1222). This remains the primary entry point. Remove the <section ROLE_TYPE="orchestrator"> block (moved to orchestrator skill) and the linter-specific resolution workflows (moved to resolver skill). Add cross-references to the two new skills.

2. holistic-linting-orchestrator — New skill ~200 lines containing the orchestrator delegation workflow currently in lines 49-245. This is the content inside <section ROLE_TYPE="orchestrator"> including: agent delegation patterns, step-by-step workflow, anti-patterns to avoid, workflow summary.

3. holistic-linting-resolver — New skill ~500 lines containing linter-specific resolution workflows currently in lines 376-991 (Ruff 380-513, Mypy 514-707, Pyright 709-957, Integration 958-991). Deduplicate the shared suppression gate sections (duplicated 3x), "Load python3-development skill" steps, and verification patterns into a common methodology section.

Additional fixes during split:

• Lines 1005-1011: Change python holistic-linting/scripts/install-agents.py to uv run ./scripts/install-agents.py for path consistency

Acceptance Criteria:

1. Main holistic-linting/SKILL.md is under 450 lines and contains only router/core content
2. holistic-linting-orchestrator/SKILL.md exists with ~200 lines of orchestrator-specific content
3. holistic-linting-resolver/SKILL.md exists with under 520 lines, deduplication applied
4. All three skills have valid YAML frontmatter with appropriate descriptions containing trigger keywords
5. Cross-references between the three skills use correct relative paths
6. No content is lost — all 1221 lines of original content are preserved across the three files (minus deduplication)
7. The install-agents.py path on former lines 1005-1011 uses uv run ./scripts/install-agents.py

Required Inputs:

• Design spec section: "Skill Split"
• Source files: ./plugins/holistic-linting/skills/holistic-linting/SKILL.md

Expected Outputs:

• ./plugins/holistic-linting/skills/holistic-linting/SKILL.md (trimmed)
• ./plugins/holistic-linting/skills/holistic-linting-orchestrator/SKILL.md (new)
• ./plugins/holistic-linting/skills/holistic-linting-resolver/SKILL.md (new)

Can Parallelize With: Task 2 Reason: Task 1 modifies skill files; Task 2 modifies an agent file — no shared files

Verification Steps:

1. Run wc -l ./plugins/holistic-linting/skills/holistic-linting/SKILL.md — under 450 lines
2. Run wc -l ./plugins/holistic-linting/skills/holistic-linting-orchestrator/SKILL.md — approximately 200 lines
3. Run wc -l ./plugins/holistic-linting/skills/holistic-linting-resolver/SKILL.md — under 520 lines
4. Verify frontmatter in all three files: head -5 ./plugins/holistic-linting/skills/*/SKILL.md
5. Grep for cross-references: grep -r "holistic-linting-orchestrator\|holistic-linting-resolver" ./plugins/holistic-linting/skills/holistic-linting/SKILL.md
6. Verify no suppression gate duplication: grep -c "Suppression Gate" ./plugins/holistic-linting/skills/holistic-linting-resolver/SKILL.md — should be 1 (shared section)
7. Verify install path fix: grep "install-agents" ./plugins/holistic-linting/skills/holistic-linting/SKILL.md — should show uv run ./scripts/

Task 2: Optimize post-linting-architecture-reviewer agent

Status: ❌ NOT STARTED Dependencies: None Priority: 2 Complexity: Medium Agent: subagent-refactorer

Target: ./plugins/holistic-linting/agents/post-linting-architecture-reviewer.md Issue Type: AGENT_OPTIMIZE

Description:

Optimize the post-linting-architecture-reviewer agent:

1. Change model from haiku to inherit (line 4). Architectural review requires reliable reasoning about design patterns, type safety, and code organization. Haiku has documented ~50% hallucination rate for reasoning tasks (per project CLAUDE.md).

2. Enhance description with additional trigger keywords: Add "design principles", "type safety validation", "SOLID compliance", "code organization" to improve delegation matching accuracy.

3. Fix markdown structure: The code fence block starting at line 116 (````markdown) has structural issues. Lines 164-165 close the inner fences but line 182 has a stray closing ```. Clean up the fence nesting so the file renders correctly.

Acceptance Criteria:

1. Frontmatter model field is inherit (not haiku)
2. Frontmatter description contains keywords: "design principles", "type safety", "code organization"
3. No stray/unclosed markdown code fences in the file
4. File renders correctly as markdown (no broken fence blocks)
5. All existing review checklist items (lines 54-111) preserved unchanged

Required Inputs:

• Design spec section: "Agent Optimizations"
• Source files: ./plugins/holistic-linting/agents/post-linting-architecture-reviewer.md

Expected Outputs:

• ./plugins/holistic-linting/agents/post-linting-architecture-reviewer.md (modified)

Can Parallelize With: Task 1 Reason: Modifies agent file only; Task 1 modifies skill files — no shared files

Verification Steps:

1. Run head -6 ./plugins/holistic-linting/agents/post-linting-architecture-reviewer.md — verify model: inherit
2. Grep description for new keywords: grep -i "design principles\|type safety\|code organization" ./plugins/holistic-linting/agents/post-linting-architecture-reviewer.md
3. Validate markdown fence balance: Count opening and closing fences match
4. Verify review checklist preserved: grep -c "\- \[ \]" ./plugins/holistic-linting/agents/post-linting-architecture-reviewer.md — should be >=20

Task 3: Update plugin.json with new skills and metadata

Status: ❌ NOT STARTED Dependencies: Task 1 Priority: 2 Complexity: Low Agent: claude-context-optimizer

Target: ./plugins/holistic-linting/.claude-plugin/plugin.json Issue Type: STRUCTURE_FIX

Description:

Update plugin.json to:

1. Add the two new skill directories to the skills array
2. Add keywords field for marketplace discoverability
3. Bump version from 1.0.7 to 1.1.0 (minor version for new skills)

Acceptance Criteria:

1. skills array contains all 3 skill paths: ["./skills/holistic-linting", "./skills/holistic-linting-orchestrator", "./skills/holistic-linting-resolver"]
2. keywords field present with relevant terms: linting, code-quality, ruff, mypy, bandit, python, formatting, type-checking
3. version bumped to "1.1.0"
4. File is valid JSON (parseable by python3 -m json.tool)

Required Inputs:

• Design spec section: "Documentation Improvements" items 2 and 3
• Source files: ./plugins/holistic-linting/.claude-plugin/plugin.json

Expected Outputs:

• ./plugins/holistic-linting/.claude-plugin/plugin.json (modified)

Can Parallelize With: None (depends on Task 1) Reason: Needs Task 1 complete to reference correct skill directories

Verification Steps:

1. Run python3 -m json.tool ./plugins/holistic-linting/.claude-plugin/plugin.json — valid JSON
2. Verify skills array: python3 -c "import json; d=json.load(open('./plugins/holistic-linting/.claude-plugin/plugin.json')); print(len(d['skills']))" — should be 3
3. Verify keywords present: grep keywords ./plugins/holistic-linting/.claude-plugin/plugin.json
4. Verify version: grep version ./plugins/holistic-linting/.claude-plugin/plugin.json — should show 1.1.0

Task V1: Validate plugin structure

Status: ❌ NOT STARTED Dependencies: Task 1, Task 2, Task 3 Priority: 1 Complexity: Low Agent: plugin-assessor

Target: ./plugins/holistic-linting/ Issue Type: STRUCTURE_FIX

Acceptance Criteria:

1. Plugin passes structural validation with no critical issues
2. All internal links resolve correctly (skills reference each other, agent references resolve)
3. No orphaned files (status maintained from baseline — 0 orphans)
4. Frontmatter validates against Claude Code schema for all 3 skills and 2 agents
5. Score improved from 68/100 baseline

Required Inputs:

• Source files: ./plugins/holistic-linting/ (entire plugin directory)

Expected Outputs:

• Validation report with updated score

Can Parallelize With: None Reason: Depends on all modification tasks completing

Verification Steps:

1. Run plugin-assessor on ./plugins/holistic-linting/ — no critical issues
2. Verify all 3 skill SKILL.md files exist and have valid frontmatter
3. Verify both agent files have valid frontmatter
4. Check all markdown links resolve: grep for ](./ in all SKILL.md files and verify targets exist
5. Score shows improvement from 68/100

Task V2: Update plugin documentation

Status: ❌ NOT STARTED Dependencies: Task V1 Priority: 2 Complexity: Low Agent: plugin-docs-writer

Target: ./plugins/holistic-linting/README.md Issue Type: DOC_IMPROVE

Acceptance Criteria:

1. README.md reflects current plugin structure with 3 skills
2. All skills listed with descriptions matching their frontmatter
3. Both agents documented with their roles
4. The /lint command documented with usage examples
5. Installation instructions accurate

Required Inputs:

• Source files: ./plugins/holistic-linting/ (entire plugin directory)
• Validation results from Task V1

Expected Outputs:

• ./plugins/holistic-linting/README.md (updated)

Can Parallelize With: None Reason: Depends on Task V1 (needs validated structure to document accurately)

Verification Steps:

1. README.md exists and has content
2. All 3 skill names appear in README: grep -c "holistic-linting-orchestrator\|holistic-linting-resolver\|holistic-linting" ./plugins/holistic-linting/README.md
3. Both agent names appear: grep -c "linting-root-cause-resolver\|post-linting-architecture-reviewer" ./plugins/holistic-linting/README.md
4. /lint command documented: grep "/lint" ./plugins/holistic-linting/README.md

Dependency Summary

Phase 1 (Parallel):
  Task 1: Split SKILL.md ──────────┐
  Task 2: Optimize agent ──────────┤
                                   │
Phase 2 (Sequential):              │
  Task 3: Update plugin.json ◄─────┘ (depends on Task 1)
                                   │
Phase 3 (Sequential):              │
  Task V1: Validate structure ◄────┘ (depends on Tasks 1, 2, 3)
                                   │
Phase 4 (Sequential):              │
  Task V2: Update README ◄────────┘ (depends on Task V1)

Parallelization Groups

• Group A (can launch simultaneously): Task 1, Task 2 — No shared files
• Group B (sequential after Group A): Task 3 — Depends on Task 1
• Group C (sequential after all modifications): Task V1 — Depends on Tasks 1, 2, 3
• Group D (sequential after validation): Task V2 — Depends on Task V1

Success Metrics

Context Manifest

Domain Boundaries in SKILL.md

The source SKILL.md (1221 lines) contains 4 distinct domains that will be split across 3 skills:

Deduplication Opportunity: Lines 376-991 contain ~36 lines of duplicated suppression gate sections (12 lines × 3 linters) and shared methodology steps. Extract common resolution pattern into shared section to reduce from ~616 lines to target ~500 lines.

Path Fix: Lines 1005-1011 use python holistic-linting/scripts/install-agents.py which is relative to skills directory parent. Will be changed to uv run ./scripts/install-agents.py for consistency with other script references.

Cross-Reference Map

After Split Impact: The main holistic-linting skill remains the entry point. Agent references Skill(command: "holistic-linting") will continue to work as the router skill will reference or incorporate resolver content. No agent file changes needed for skill references.

External Dependencies

External files that reference holistic-linting (refactoring will NOT break these):

Assessment: No external file changes required. All references are either to the plugin name (unchanged) or generic mentions of "holistic-linting skill" which remain valid as the main skill persists as the entry point.

Reference Index Structure

The rules knowledge base uses a three-tier progressive disclosure hierarchy:

Tier 1: SKILL.md (lines 1021-1097) — Documents existence of rules knowledge base with summary tables:

• Ruff: 933 rules, 19 families documented
• Mypy: Error codes organized by category
• Bandit: 65+ security checks organized by vulnerability type

Tier 2: Index Files (3 files) — Provide rule family/category organization and navigation:

• ./references/rules/ruff/index.md — Lists 19 rule families (pycodestyle, pyflakes, bugbear, etc.) with links to family-specific files
• ./references/rules/mypy/index.md — Organizes error codes by category (attribute access, function calls, etc.) with links to error documentation
• ./references/rules/bandit/index.md — Organizes security checks by vulnerability category (credentials, crypto, injection, etc.)

Tier 3: Detailed Rule Files (52 files total in ./references/rules/{ruff,mypy,bandit}/) — Individual rule documentation with examples, principles, and resolution patterns

Link Patterns: All links use relative markdown paths starting with ./ from the referencing file's location. SKILL.md → ./references/rules/ruff/index.md, index files → ./pycodestyle-errors-warnings.md

After Split: The router skill (holistic-linting) will retain documentation of the rules knowledge base (lines 1021-1097). The resolver skill (holistic-linting-resolver) will reference the rules KB but does NOT need to duplicate index links — agents activate the main skill which provides navigation to the KB.

Additional Reference Files

Beyond the rules knowledge base, the plugin contains:

Mypy Cached Documentation (6 .rst files in ./references/mypy-docs/):

• error_code_list.rst and error_code_list2.rst — Referenced in SKILL.md lines 526-529 as locally-cached mypy error code documentation
• Used by Mypy Resolution Workflow for offline rule lookup
• Paths: ./references/mypy-docs/error_code_list.rst (relative from SKILL.md)

Online Resource Access Guide:

• ./references/accessing_online_resources.md — Referenced in SKILL.md line 749 for MCP tool usage guidance when researching Pyright rules
• Provides instructions for using mcp__Ref__ref_search_documentation, mcp__exa__get_code_context_exa, WebFetch tools

Scripts:

• ./scripts/detect-hook-tool.py — Git hook tool detection (prek vs pre-commit)
• ./scripts/install-agents.py — Agent installation to user/project scope
• Referenced in SKILL.md lines 302-310 (detection) and 1005-1011 (installation)

All reference files use relative paths from SKILL.md location and remain unchanged by the split.

Key Observations

Architectural Pattern: The current SKILL.md functions as a "kitchen sink" containing:

1. Router logic (what skill does, when to use, linter detection)
2. Role-specific workflows (orchestrator delegation inside XML section tag)
3. Execution procedures (linter-specific resolution workflows for sub-agents)
4. Resource documentation (bundled agents, rules KB, scripts)

Split Rationale: The XML section tag on line 49 (<section ROLE_TYPE="orchestrator">) explicitly delineates orchestrator-only content, suggesting original intent to separate concerns. The refactoring actualizes this separation into discrete skills.

Deduplication Targets:

• Suppression Gate section: Lines 458-470 (Ruff), 598-610 (Mypy), 825-837 (Pyright) — Identical text duplicated 3 times
• "Load python3-development skill" step: Lines 449-456 (Ruff), 589-596 (Mypy), 816-823 (Pyright) — Near-identical with slight wording variations
• Verification step: Lines 483-488 (Ruff), 668-674 (Mypy), 911-919 (Pyright) — Similar pattern with tool name variation

Extract common methodology once, then linter-specific sections reference it with tool-specific variations.

Agent Model Issue: post-linting-architecture-reviewer.md line 4 sets model: haiku. Architectural review requires reasoning about design patterns, code organization, type safety, SOLID principles — tasks where Haiku has documented ~50% hallucination rate per project CLAUDE.md. Model should be inherit to use orchestrator's reasoning-capable model.

Markdown Structure Issue: post-linting-architecture-reviewer.md lines 116-182 have fence nesting problems. Opening `````markdown` on line 116, but closing structure is malformed with stray backticks on lines 165, 182. This prevents proper markdown rendering.

No Orphaned Files: All 52 reference files in ./references/rules/ are indexed through the three-tier hierarchy. The 6 mypy-docs RST files are explicitly referenced in SKILL.md. The accessing_online_resources.md is linked inline. Scripts are documented in bundled resources section. Assessment confirmed: 0 orphaned files.

Version Bump Required: plugin.json version is 1.0.7. Adding two new skills constitutes a minor feature addition. Version should bump to 1.1.0 per repository version bumping rules.

Cross-Plugin Integration Points:

• python3-development skill: Activated in all 3 linter resolution workflows (lines 453, 593, 821) to ensure Python 3.11+ standards
• agent-orchestration skill: References holistic-linting workflows in delegation templates
• verification-gate skill: May reference linting validation in completion gates

These integration points use skill activation syntax (Skill(command: "holistic-linting")) which continues to work as the main skill name is unchanged.