witch-delegation
Use after design and plan approval to execute a plan through Witch's fixed work and review lanes
Skill by muratmirgun
Install
Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.
npx agentmarket install skill/muratmirgun-witch-delegation
Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.
Get the weekly diff
The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.
About
Use cases: design
Reviews
Write a reviewNo reviews yet. Installed it? Reviews come from verified installers only. How reviews work
Permissions
What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.
- Tools
- No tools found
- Secrets
- None asked for
- Install scripts
- None found
Network hosts it talks to 1
codeload.github.com· found in install command
Files it touches
~/.claude/skills/witch-delegation
What's inside
- Skill name
witch-delegation- Description
- Use after design and plan approval to execute a plan through Witch's fixed work and review lanes
- License
- Apache-2.0
- Instructions
- 30 lines in SKILL.md · view on GitHub
Files (3)
- SKILL.md
- dispatch-envelope.md · references/
- review-contracts.md · references/
SKILL.md (first lines)
# Witch Delegation Execute approved work through fixed Witch-owned lanes. The controller owns routing, dispatch context, verification, the repair ledger, and the decision to ship. ## Fixed lanes Work lanes are `witch-routine`, `witch-complex`, `witch-researcher`, and `witch-security`. Review lanes are `witch-task-reviewer`, `witch-re-reviewer`, and `witch-final-reviewer`. Never substitute a generic subagent for a Witch lane. Do not override the selected agent profile's model or reasoning effort in a dispatch. Do not automatically promote work to another lane after a failure or an unfavorable review. ## Ordered procedure 1. Begin only with an approved design and an approved plan. 2. Decompose the plan into bounded tasks with explicit ownership and proof. 3. Call `witch_route` before every implementation or research work dispatch. Record the route output and decisive reason. Security-sensitive work routes to `witch-security`; research that does not mutate the worktree routes to `witch-researcher`; all other work follows the tool's fixed selection between `witch-routine` and `witch-complex`. 4. Fill every section of the [Dispatch Envelope](references/dispatch-envelope.md), then dispatch the selected work lane without model, reasoning-effort, or permission overrides. 5. Verify the returned diff, owned-file boundaries, and test evidence as controller. Do not treat an implementer's report as proof. 6. Dispatch a fresh `witch-task-reviewer` using the closed [Review Contracts](references/review-contracts.md). 7. On `repair`, send the bounded findings to the same implementation lane, then dispatch a fresh `witch-re-reviewer`. Repeat only within the repair breaker. Never automatically promote the repair to a different work lane. 8. When the task verdict is accepted or all load-bearing findings are closed, record the task, route, evidence, review verdicts, repairs, and residual risks in the ledger. 9. After every task is complete, dispatch a fresh `witch-final-reviewer` over the whole change set. Final-review repairs return to their original implementation lane and require a new fresh final review. 10. Ship only on `FINAL VERDICT: ship`. Stop when the repair breaker trips, when approval is required, or when the final verdict is `redesign`. The complete sequence is: approved design → approved plan → task decomposition → `witch_route` → Dispatch Envelope → selected work lane → controller diff/test verification → `witch-task-reviewer` → bounded same-lane repair + `witch-re-reviewer` → ledger completion → fresh `witch-final-reviewer` → ship. ## Breaker Two unsuccessful repair rounds for the same load-bearing finding trip the breaker. Stop delegation, preserve the evidence and open finding in the ledger, and report the block to the user. Do not route again or change lanes as an automatic escalation.
From muratmirgun/witch, Apache-2.0.
More from muratmirgun/witch
- 0.553score
using-witch
muratmirgun
Use at the start of every Witch session to select the governing process skill and preserve Witch routing boundaries
SkillBFreshness: B★ 3Claude Code +9 - 0.543score
subagent-driven-development
muratmirgun
Use when executing implementation plans with independent tasks in the current session
SkillBFreshness: B★ 3Claude Code +9 - 0.538score
systematic-debugging
muratmirgun
Use when encountering any bug, test failure, or unexpected behavior, before proposing fixes
SkillBFreshness: B★ 3Claude Code +9 - 0.538score
executing-plans
muratmirgun
Use when you have a written implementation plan to execute in a separate session with review checkpoints
SkillBFreshness: B★ 3Claude Code +9 - 0.538score
writing-plans
muratmirgun
Use when you have a spec or requirements for a multi-step task, before touching code
SkillBFreshness: B★ 3Claude Code +9 - 0.538score
requesting-code-review
muratmirgun
Use when completing tasks, implementing major features, or before merging to verify work meets requirements
SkillBFreshness: B★ 3Claude Code +9 - 0.536score
brainstorming
muratmirgun
You MUST use this before any creative work - creating features, building components, adding functionality, or modifying behavior.
SkillBFreshness: B★ 3Claude Code +9
Signals
- Stars
- 3
- Forks
- 0
- Open issues
- 0
- Last commit (this folder)
- Aug 2, 2026 (2 months ago)
- Last release
- Aug 15, 2026 (2 months ago)
From GitHub (fetched Oct 24, 2026).
Details
- License
- Apache-2.0
- Repository
- github.com/muratmirgun/witch
- First seen
- Oct 24, 2026
- Last checked
- Oct 24, 2026
- Source
- GitHubgithub-graphql
More from muratmirgun
- 0.553score
using-witch
muratmirgun
Use at the start of every Witch session to select the governing process skill and preserve Witch routing boundaries
SkillBFreshness: B★ 3Claude Code +9 - 0.543score
subagent-driven-development
muratmirgun
Use when executing implementation plans with independent tasks in the current session
SkillBFreshness: B★ 3Claude Code +9 - 0.538score
systematic-debugging
muratmirgun
Use when encountering any bug, test failure, or unexpected behavior, before proposing fixes
SkillBFreshness: B★ 3Claude Code +9 - 0.538score
executing-plans
muratmirgun
Use when you have a written implementation plan to execute in a separate session with review checkpoints
SkillBFreshness: B★ 3Claude Code +9 - 0.538score
writing-plans
muratmirgun
Use when you have a spec or requirements for a multi-step task, before touching code
SkillBFreshness: B★ 3Claude Code +9 - 0.538score
requesting-code-review
muratmirgun
Use when completing tasks, implementing major features, or before merging to verify work meets requirements
SkillBFreshness: B★ 3Claude Code +9
Cite this page
witch-delegation skill. Agent Mart. https://agentmart.dev/skills/muratmirgun-witch-delegation (last updated 2026-10-24)