Delegate Skill — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited Delegate Skill (Plugin) and scored it 96/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 1 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 1 flagged
The text {match} tells the agent to skip the normal "ask the user first" gate. Used adversarially it removes the human-in-the-loop check before destructive or sensitive actions, turning a normally-gated agent into a fire-and-forget executor.
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
Turn one task into five parallel agents. Describe what you want built, get a structured report when it's done.
Delegate is a Claude Code skill that breaks implementation tasks into independent work units and executes them in parallel. Instead of one agent working through files sequentially, Delegate decomposes the task, writes self-contained specs, spawns up to 5 concurrent agents, reviews the results, and reports back.
It handles orchestration so you stay at the architecture level — describe the feature, review the output.
Delegate runs in 6 phases:
Your task description
|
[1. Explore] -- understand codebase
|
[2. Decompose] -- break into units
|
[3. Write Specs] -- self-contained instructions
|
[4. Execute] -- spawn parallel agents
/ | \
[A] [B] [C] -- up to 5 concurrent
\ | /
[5. Review] -- verify & fix integration
|
[6. Report] -- structured summaryDelegate gets better the more detailed your plan is.
This is the most important thing to understand. Delegate's Phase 1 (Explore) reads your codebase to understand patterns, but it can't read your mind. The more context you provide upfront, the better the specs it writes, and the better the output from every agent.
/delegate Add user authentication/delegate Add JWT authentication using the existing auth middleware pattern
in src/middleware/auth.ts. Add login/register endpoints to src/routes/auth.ts,
a User model in src/models/user.ts, and update the route index to include
auth routes./delegate Implement the auth system from our plan:
- JWT tokens with refresh rotation (see src/middleware/auth.ts pattern)
- User model with bcrypt password hashing (follow src/models/post.ts structure)
- Login/register/refresh endpoints in src/routes/auth.ts
- Protect all /api/v2/* routes with requireAuth middleware
- Add auth error handling to src/middleware/errorHandler.tsA detailed plan means Delegate writes better specs. Better specs mean agents produce code that matches your codebase on the first try. First-try accuracy means less review fixing and faster completion.
Pro tip: Use Claude Code's plan mode first (/plan), get the approach right, then hand the plan to Delegate. Planning is thinking. Delegate is execution.
# Create the skill directory
mkdir -p .claude/skills/delegate
# Copy the skill file
curl -o .claude/skills/delegate/SKILL.md \
https://raw.githubusercontent.com/aayoawoyemi/Delegate-Skill/main/SKILL.md
# That's it. Use it:
# /delegate [your task description]Or manually: copy SKILL.md into .claude/skills/delegate/SKILL.md in your project.
--=={ delegate }==--
Task: Add JWT authentication with login/register endpoints
Batches: 2 (sequential -- model before routes)
Agents spawned: 4
Work Units:
1. User model + migration -- done
2. Auth middleware -- done
3. Auth routes (login/register/refresh) -- done
4. Route index + error handler updates -- done
Files Created:
- src/models/user.ts -- User model with bcrypt hashing
- src/routes/auth.ts -- login, register, refresh endpoints
- src/migrations/003_users.sql -- users table migration
Files Modified:
- src/routes/index.ts -- added auth route imports
- src/middleware/errorHandler.ts -- added AuthError handling
- src/middleware/auth.ts -- added requireAuth and optionalAuth
Integration Fixes: 1 (aligned User type import between model and routes)
Retries: none
Ready to test: npm run dev, then POST /api/auth/register with { email, password }| Manual Claude Code | With Delegate | |
|---|---|---|
| Multi-file tasks | Sequential, one file at a time | Up to 5 files in parallel |
| Agent context | Grows with each file, quality degrades | Fresh context per agent |
| File conflicts | Possible if you're not careful | Worktree isolation (code) or non-overlap rules (other) |
| Failure recovery | You debug and retry | Auto-retry once, then self-fix |
| You manage | Every file edit | Just the task description |
MIT
Built by Aayo Awoyemi. Inspired by patterns from Composio's agent-orchestrator (fleet isolation, declarative reactions) adapted for single-session Claude Code workflows.
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.