⚠️ Archived — moved into the delego plugin. These agents now live inDelego-Dev/plugin(the single source of truth). Install everything in one command:/plugin marketplace add Delego-Dev/plugin /plugin install delego@delegoOr à la carte: clone the plugin repo and run its
install.sh. The content below is a frozen snapshot.
Claude Code subagents for working with delego — the policy & audit firewall for agent actions. Agents are the autonomous review surface: focused reviewers and investigators you delegate to. They're clustered by function.
# in your project: clone into the folder Claude Code reads (recursive discovery)
git clone https://github.com/Delego-Dev/agents .claude/agents/delegoRestart Claude Code, then delegate by name — e.g.
@agent-delego-policy-reviewer review .claude/.delego/policy.yaml.
| Cluster | Agent | Use it to… |
|---|---|---|
review/ |
delego-policy-reviewer |
Adversarially review a policy.yaml — fail-closed correctness, over-broad rules, missing forbidden rules, constraint gaps, and the known footguns; returns severity-rated findings. |
review/ |
delego-broker-reviewer |
Review a custom BrokerAdapter for the invariants — no credential in delego's process, executes exactly the authorised action, fails closed. |
audit/ |
delego-audit-investigator |
Verify the signed ledger, reconstruct an action's authority path, and flag anomalies (substituted-action refusals, single-use replays, denial clusters). |
For the interactive surface (init, policy drafting, approval triage, audit explaining) see the companion skills repo.
Claude Code discovers subagents recursively under .claude/agents/, so the
cluster layout (review/, audit/) is preserved — clone the repo into a subdir
of .claude/agents/ (safe even if you already have other agents):
cd your-project
git clone https://github.com/Delego-Dev/agents .claude/agents/delegoThen restart Claude Code (subagents load at session start). Update later with
git -C .claude/agents/delego pull (and restart). The repo's README/scripts/
have no agent frontmatter, so discovery ignores them.
- By name (guaranteed):
@agent-delego-policy-reviewer - Natural language: "use the delego-policy-reviewer agent to review
policy.yaml" - Claude also auto-delegates when your request matches an agent's
description.
@agent-delego-policy-reviewer review .claude/.delego/policy.yaml
use the delego-broker-reviewer agent on my broker.py
@agent-delego-audit-investigator verify the ledger and flag anything odd
Draft with the delego-policy-drafter skill → harden with the
delego-policy-reviewer agent. Operate with the delego-approval-triage /
delego-audit-explainer skills → investigate incidents with the
delego-audit-investigator agent.
Each agent is a markdown file under a function cluster (<cluster>/<name>.md) with
YAML frontmatter: a kebab-case name (unique across the repo — recursive
discovery means a duplicate silently wins), a description that says when to
delegate to it, and optional tools / model. The body is the agent's system
prompt. python scripts/validate.py (run in CI) checks every agent. New clusters
welcome (e.g. simulate/ for adversarial policy testing, respond/ for incident
response) — fork and open a PR.
Apache-2.0. Built for delego; see the wire specification.