subagent-tweaker
Fix, improve, or update existing Claude Code custom agents (subagents). Use when: (1) User reports an agent isn't working well, (2) User wants to adjust agent behavior, tools, or model, (3) User says 'fix agent', 'update agent', 'tweak agent', 'agent not working'. Edits agent frontmatter, tool restrictions, prompts.
Best use case
subagent-tweaker is best used when you need a repeatable AI agent workflow instead of a one-off prompt.
Fix, improve, or update existing Claude Code custom agents (subagents). Use when: (1) User reports an agent isn't working well, (2) User wants to adjust agent behavior, tools, or model, (3) User says 'fix agent', 'update agent', 'tweak agent', 'agent not working'. Edits agent frontmatter, tool restrictions, prompts.
Teams using subagent-tweaker should expect a more consistent output, faster repeated execution, less prompt rewriting.
When to use this skill
- You want a reusable workflow that can be run more than once with consistent structure.
When not to use this skill
- You only need a quick one-off answer and do not need a reusable workflow.
- You cannot install or maintain the underlying files, dependencies, or repository context.
Installation
Claude Code / Cursor / Codex
Manual Installation
- Download SKILL.md from GitHub
- Place it in
.claude/skills/subagent-tweaker/SKILL.mdinside your project - Restart your AI agent — it will auto-discover the skill
How subagent-tweaker Compares
| Feature / Agent | subagent-tweaker | Standard Approach |
|---|---|---|
| Platform Support | Not specified | Limited / Varies |
| Context Awareness | High | Baseline |
| Installation Complexity | Unknown | N/A |
Frequently Asked Questions
What does this skill do?
Fix, improve, or update existing Claude Code custom agents (subagents). Use when: (1) User reports an agent isn't working well, (2) User wants to adjust agent behavior, tools, or model, (3) User says 'fix agent', 'update agent', 'tweak agent', 'agent not working'. Edits agent frontmatter, tool restrictions, prompts.
Where can I find the source code?
You can find the source code on GitHub using the link provided at the top of the page.
Related Guides
SKILL.md Source
# Tweak Custom Agent ## Workflow ### Step 1: Identify the agent Find agents at: - `$HOME/.claude/agents/*.md` (personal) - `.claude/agents/*.md` (project) Read the agent file to understand current configuration. ### Step 2: Diagnose the issue Common problems and fixes: | Problem | Likely Cause | Fix | |---------|-------------|-----| | Agent not being used | Poor `description` | Rewrite with clear trigger keywords | | Agent too slow | Wrong `model` | Switch to `sonnet` or `haiku` | | Agent can't edit files | `tools` missing Write/Edit | Add needed tools to allowlist | | Agent doing too much | No tool restrictions | Add `tools:` or `disallowedTools:` | | Agent forgets context | No persistent memory | Add `memory: user` or `memory: project` | | Agent runs too long | No turn limit | Add `maxTurns: N` | ### Step 3: Apply changes Edit the agent's Markdown file. Preserve existing content that works well. **Frontmatter fields reference:** | Field | Description | |-------|-------------| | `name` | Identifier (lowercase, hyphens) | | `description` | When to use (Claude reads this for delegation) | | `model` | `opus`, `sonnet`, `haiku`, `inherit` | | `tools` | Tool allowlist (inherits all if omitted) | | `disallowedTools` | Tool denylist | | `permissionMode` | `default`, `acceptEdits`, `delegate`, `dontAsk`, `bypassPermissions`, `plan` | | `maxTurns` | Max agentic turns | | `skills` | Preloaded skills | | `mcpServers` | Available MCP servers | | `hooks` | Scoped lifecycle hooks | | `memory` | `user`, `project`, `local` | ### Step 4: Format the agent file Format the edited agent file using the mdx-formatter to ensure consistent markdown formatting: ```bash pnpm dlx @takazudo/mdx-formatter --write <path-to-agent-file.md> ``` ### Step 5: Verify After editing: 1. Confirm YAML frontmatter has no syntax errors 2. Check tool restrictions match intended behavior 3. Verify description contains keywords users would say ## Tips - Keep agent body (system prompt) focused and concise - Subagents cannot nest - don't add `Task` tool to agents that will be spawned as subagents - When adjusting tools, prefer `disallowedTools` over `tools` if only blocking a few - `description` is the primary trigger mechanism - this is the most impactful field to improve
Related Skills
subagent-creator
Create new Claude Code custom agents (subagents). Use when: (1) User wants to create a new custom agent, (2) User says 'create agent', 'new agent', 'make subagent', (3) User wants a specialized agent for delegation. Covers agent file format, YAML frontmatter, tool restrictions, model selection, permission modes, persistent memory, placement.
zudoesa-articlify
Convert conversation context into an esa article via the zudoesa-writer subagent. ONLY invoke when the user explicitly asks — NEVER proactively propose. Triggers: 'write esa article', 'esa記事', 'esaに書いて', 'articlify for esa', or /zudoesa-articlify. Gathers context, creates a writing brief, delegates to the writer subagent.
zudoesa-apply-voice
Apply Takazudo's esa writing voice and vocabulary rules to text. Use when: (1) User wants to write/rewrite text in Takazudo's esa style, (2) User says 'apply voice', 'esa voice', 'esa文体で', 'esa風に書いて', '文体を適用', (3) User provides text to transform to esa style. Reads writing-style.md and vocabulary-rule.md from takazudo-esa-writing repo and applies the rules.
zudocg-articlify
Convert conversation context into a CodeGrid article via the zudocg-writer subagent. ONLY invoke when the user explicitly asks — NEVER proactively propose. Triggers: 'write codegrid article', 'CodeGrid記事', 'codegridに書いて', 'articlify for codegrid', or /zudocg-articlify. Gathers context, creates a writing brief, delegates to the writer subagent.
zudocg-apply-voice
Apply Takazudo's CodeGrid writing voice and vocabulary rules to text. Use when: (1) User wants to write/rewrite text in Takazudo's CodeGrid style, (2) User says 'apply voice', 'codegrid voice', 'codegrid文体で', 'codegrid風に書いて', '文体を適用', (3) User provides text to transform to CodeGrid style. Reads writing-style.md and vocabulary-rule.md from takazudo-codegrid-writing repo and applies the rules.
zpaper-articlify
Convert conversation context into a zpaper blog article via the zpaper-writer subagent. ONLY invoke when the user explicitly asks — NEVER proactively propose. Triggers: 'write zpaper article', 'zpaper記事', 'zpaperに書いて', 'articlify for zpaper', or /zpaper-articlify. Gathers context, creates a writing brief, delegates to the writer subagent.
zpaper-apply-voice
Apply Takazudo's zpaper blog writing voice and vocabulary rules to text. Use when: (1) User wants to write/rewrite text in Takazudo's zpaper style, (2) User says 'apply voice', 'zpaper voice', 'zpaper文体で', 'zpaper風に書いて', 'ブログ文体を適用', (3) User provides text to transform to zpaper style. Reads writing-style.md and vocabulary-rule.md from the zpaper repo and applies the rules.
xlsx
Spreadsheet creation, editing, and analysis. Use when working with .xlsx, .xlsm, .csv, .tsv files for: (1) Creating spreadsheets with formulas and formatting, (2) Reading or analyzing data, (3) Modifying existing spreadsheets while preserving formulas, (4) Data analysis and visualization, (5) Recalculating formulas.
x
Facade for development workflows. Routes on two axes: plan-first vs implement-now (escalates to /big-plan -a when the request needs research / decomposition / has unclear scope — the appended -a makes the plan chain into implementation in-session), then single vs multi on the ready-to-build fast paths (/x-as-pr single-topic, /x-wt-teams multi-topic parallel). Use when: (1) User says '/x' followed by dev instructions, (2) User wants to start development without choosing the workflow skill, (3) User says 'dev', 'implement', or 'build' with a task. Default option: -v (verify-ui). Review-loop (-l) is opt-in — without -l the downstream skill runs a single /deep-review pass. Forwards -a (autonomy/auto-chain) and -m (merge at the end + cleanup + CI watch) through every route; auto-fix of raised findings (-f) and issue-raising (-ri) are downstream defaults, with -nf/--no-fix and -nori/--no-raise-issues as the forwarded opt-outs. -a and -m are orthogonal — full hands-off end-to-end is -a -m.
x-wt-teams
Parallel multi-topic development using git worktrees, base branches, and Claude Code agent teams. Use when: (1) User wants to work on multiple related features in parallel, (2) User mentions 'worktree', 'base branch', 'parallel development', 'split into topics', or 'multi-topic'. FULLY AUTONOMOUS — creates worktrees, spawns teams, coordinates everything. Also supports Super-Epic child mode for [Epic] issues from /big-plan with '**Super-epic:** #N' markers (targets the super-epic base branch instead of main).
x-as-pr
Start a development workflow as a draft PR. Creates a NEW branch from the current branch, empty start commit, draft PR targeting the current branch, then implements. ALWAYS creates a new branch by default — produces a nested PR-on-PR when the current branch already has one. Use when: (1) User says 'dev as pr', (2) User wants a PR-first workflow before coding, (3) User passes -s/--stay to reuse the current branch instead of nesting, (4) User passes a GitHub issue URL to implement, (5) User passes --make-issue/--issue to create an issue first. Logs progress via issue comments when an issue is linked.
watch-ci
Watch GitHub PR CI checks in the background and notify on completion. Use when: (1) User wants to monitor CI/CD status, (2) User says 'watch CI', 'check CI', 'monitor checks', or 'wait for CI', (3) User wants to know when checks pass or fail. Runs a background gh polling shell loop (NOT a subagent — near-zero token cost), sends macOS notification on completion. Also handles merged PRs by watching the target branch CI.