skill-maintenance
Produces a maintained skills directory with updated names, unified terminology, repaired cross-references, and a git commit documenting every change and what was preserved. Use when: 'update these skill names', 'refactor the skills directory', 'clean up skill references', 'rename this skill', 'audit the skills ecosystem', 'deprecate this skill', 'add version history to a skill'.
- License 许可
- Apache-2.0
- 最近更新
- 2026-10-25
- Stars
- ★ 2
- Forks
- 0
01 前言 Front Matter
- name
- skill-maintenance
- tier
- 1
- model
- sonnet
- agents
- [ "primary" ]
- inputs
- [ { "name": "skills_directory", "type": "string", "required": true, "description": "Path to the skills directory to maintain and audit" } ]
- outputs
- [ { "name": "maintenance_report", "type": "string", "description": "Maintained skills directory with updated names, unified terminology, repaired cross-references, and git commit documentation" } ]
- category
- skill-forge
- triggers
- [ "update these skill names", "refactor the skills directory", "clean up skill references", "rename this skill", "audit the skills ecosystem", "deprecate this skill" ]
- description
- Produces a maintained skills directory with updated names, unified terminology, repaired cross-references, and a git commit documenting every change and what was preserved. Use when: 'update these skill names', 'refactor the skills directory', 'clean up skill references', 'rename this skill', 'audit the skills ecosystem', 'deprecate this skill', 'add version history to a skill'.
- tool_dependencies
- [ "file_system", "bash" ]
03 SKILL.md 正文
Skill Maintenance Ritual
I. The Workflow
This is a 9-step workflow for maintaining the skills directory.
Step 1: Recognize the Need for Maintenance
Goal: Identify when skill maintenance is needed.
Triggers:
- User requests skill renames or refactors
- You notice unclear or inconsistent skill names
- New skills are added that reference outdated names
- Terminology changes in the broader ecosystem (e.g., tool names, industry standards)
- Periodic audit schedule (e.g., quarterly)
Actions:
- Pause and clarify the scope with the user
- Ask: What specifically needs to change?
- Ask: What should stay the same?
- Document the maintenance goal clearly
Output: Clear understanding of maintenance scope
Key Insight: Always pause and clarify scope before large refactors. Avoid over-refactoring by understanding what actually needs to change.
Step 2: Read Before Proposing
Goal: Understand what the skills actually do before suggesting changes.
Actions:
- Read all skills that will be affected by the maintenance
- Understand the actual purpose and workflow of each skill
- Note any cross-references between skills
- Identify patterns in naming or terminology
Tools:
filetool (read action) for each skill- Take notes on what each skill does
Output: Deep understanding of affected skills
Key Insight: Never propose renames or refactors without reading the actual content first. Names should reflect reality, not assumptions.
Step 3: Propose Clear, Descriptive Changes
Goal: Suggest changes that improve clarity and consistency.
Actions:
- For renames: Propose names following the "verb-object" pattern
- Examples:
release-specification,implementation-prompt
- Examples:
- For terminology refactors: Identify what should change and what should stay
- Create a table showing old → new with rationale
- Get user confirmation before proceeding
Tools:
messagetool (ask type) to propose and get feedback
Output: Agreed-upon changes
Key Insight: Good naming is good documentation. Descriptive names reduce cognitive load and make skills immediately understandable.
Step 4: Execute Renames Systematically
Goal: Rename skill directories and update internal references.
Actions:
- Rename skill directories using
mvcommand - Update
namefield in each SKILL.md frontmatter - Update title (H1 heading) in each SKILL.md
- Verify renames with
lscommand
Tools:
shelltool for directory renamesfiletool (edit action) for content updates
Output: Renamed skills with updated metadata
Key Insight: Rename both the directory and the internal metadata. Inconsistency between directory name and skill name causes confusion.
Step 5: Search and Catalog References
Goal: Find all instances of terminology or names that need updating.
Actions:
- Use
grep -r -i "<term>"to find all references - Count references by directory:
grep -r -i "<term>" | cut -d: -f1 | sed 's|^\./||' | cut -d/ -f1 | sort | uniq -c | sort -rn - Identify which references need updating (skills directory) vs. which should stay (historical docs)
- Confirm scope with user if needed
Tools:
shelltool with grep for searchingwc -lto count referencescut,sort,uniq -cto categorize
Output: Catalog of references to update
Key Insight: Always catalog before refactoring. Understanding the scope prevents over-refactoring or missing references.
Step 6: Read Context and Determine Strategy
Goal: Understand which references should change and which should stay.
Actions:
- Use
matchtool (grep action) to view references with context - Analyze each usage to determine if it should be updated
- Create a refactoring strategy:
- What should be replaced?
- What should be preserved?
- What replacement text should be used?
- Document the strategy
Tools:
matchtool (grep action) with leading/trailing context
Output: Refactoring strategy document
Example Strategy:
- Replace "Zenflow prompt" → "implementation prompt"
- Replace "Zenflow" (as executor) → "implementation agent"
- Preserve "Zenflow" when listing multiple tools: "Zenflow, Claude Code, etc."
- Preserve "Zenflow" in routing decisions: "Zenflow: Strategic implementation"
Key Insight: Not all references should be changed. Preserve tool-specific references when contextually appropriate.
Step 7: Execute Refactor with Batch Edits
Goal: Update all references systematically using batch edits.
Actions:
- For each affected skill file:
- Create a list of find/replace pairs
- Use
filetool (edit action) with multiple edits - Set
all: trueto replace all occurrences
- Verify changes with
grepafter each file - Keep a count of replacements per file
Tools:
filetool (edit action) with multiple editsshelltool with grep to verify
Output: Updated skill files
Example Edit:
{
"edits": [
{"all": true, "find": "Zenflow prompt", "replace": "implementation prompt"},
{"all": true, "find": "Zenflow", "replace": "implementation agent"}
]
}
Key Insight: Batch edits are more efficient than one-by-one replacements. Use all: true to replace all occurrences in a single operation.
Step 8: Verify and Commit
Goal: Ensure all changes are correct and commit with comprehensive documentation.
Actions:
- Verify no unintended references remain:
grep -i "<old term>" <directory> - Check git status:
git status - Stage changes:
git add skills/ - Write comprehensive commit message:
- Summary of changes
- File-by-file breakdown
- Rationale for changes
- What was preserved and why
- Commit:
git commit -m "<message>" - Push to remote:
git push origin main
Tools:
shelltool with git commands
Output: Committed and pushed changes
Commit Message Template:
<Action> in skills directory
<Summary paragraph>
Changes:
- <file1> (<N> replacements)
- <change 1>
- <change 2>
- <file2> (<N> replacements)
- <change 1>
Kept <term> only when:
- <context 1>
- <context 2>
<Rationale paragraph>
Key Insight: Comprehensive commit messages are documentation. Future maintainers need to understand why changes were made, not just what changed.
Step 9: Document the Process
Goal: Create summary documents for future reference.
Actions:
- Create a summary document with:
- Overview of changes
- Refactoring strategy
- Files modified
- Benefits of the changes
- Reflection on the process
- Deliver summary to user with
messagetool
Tools:
filetool (write action)messagetool (result type)
Output: Documentation for future reference
Key Insight: Document the process immediately after completing it. Details fade quickly from memory.
Output
- Renamed skill directories with updated
namefield and H1 heading in each SKILL.md - All cross-references updated to the new names or terminology
- Zero stale references remaining (verified by grep)
- A git commit with a structured message: what changed, why, what was preserved
- Optional: a summary document delivered to the user if the scope was large
Examples
Scenario 1: "Rename zenflow-prompt to implementation-prompt and update all references" → Read the skill, propose the rename following verb-object pattern, execute directory mv + metadata update, grep for all cross-references, batch-edit and verify, commit with per-file replacement counts.
Scenario 2: "Deprecate old-pipeline skill — it's been replaced by batch-normalize-and-package" → Add a deprecated: true field to frontmatter, add a deprecation notice at the top of the body pointing to the replacement, update any skills that link to it, commit.
Edge Cases
- User requests a rename but the new name is already taken by another skill — surface the conflict, propose a resolution, and get confirmation before executing
- Terminology refactor touches historical documents (retrospectives, plans) outside the skills directory — catalog them but do not change them unless the user explicitly expands the scope
- A skill has no cross-references in the rest of the directory — rename is safe; note this in the commit message
- User wants to "clean up" a skill without changing its name — treat as in-place refactoring: read, propose specific changes, get confirmation, then edit
Anti-Patterns
- Over-refactoring: Changing references that are contextually appropriate (e.g., tool-specific mentions in routing docs) because they happen to match the search term — always read the surrounding context before replacing
- Proposing without reading: Suggesting renames based on the directory name alone without reading what the skill actually does — names should reflect reality, not assumptions
- Relying on memory for references: Skipping grep and trusting recall to find all instances — always catalog systematically before refactoring
- Vague commit messages: Writing "updated skills" without a per-file breakdown — commit messages are the only audit trail for future maintainers
04 同仓库更多 Skills
kata-test-echo
暂无简介
kata-plain
An ordinary prose skill; not dispatchable.
kata-broken
暂无简介
voice-before-structure
Read the project's design language and philosophy before writing structural artifacts. Ground manifests, READMEs, and descriptions in the project's voice, not boilerplate. Use when creating manifests or READMEs for established projects, writing ecosystem-level documentation, building plugin descriptions or marketplace metadata, or authoring metadata that carries a project's identity. Trigger phrases: 'write the manifest', 'create a README', 'describe this plugin', 'ground yourself in the philosophy', 'ecosystem-level work'.
system-prompt-archaeology
Produces a named ADA disposition preset (YAML) and a layered behavioral analysis (structural decomposition + evidence-cited classifications) from a system prompt. Use when: "analyze this system prompt", "extract disposition from prompt", "reverse-engineer agent behavior", "what patterns does this prompt use", "build a preset from this prompt".
session-lifecycle-automation
Designs and configures all three loops of solo-operator session automation — session bookends, background maintenance agents, and a delegation flywheel. Use when setting up a new workspace, recovering from the 44% zero-artifact session leak, or preparing to scale delegation across multiple concurrent agents.