delegation record
A worker can return evidence. It cannot certify its own result.
suspec-agents
Claude Code worker files for Suspec roles.
Add one when a role needs its own operating context.
- 01Role
- 02Inputs
- 03Tools
- 04Evidence
note
role files
workers
6
hooks
2
home
.claude/agents
This page is a roster. The repo holds the files; the docs explain when to delegate.
lane 01 — read-only
Read-only lane
Review, inspect, verify, or challenge without writing artifacts.
lane 02 — write one artifact
Bounded-authoring lane
Draft one named artifact. Review still decides what it means.
install.sh — copy one worker
# copy one worker
$ REPO=<your-repo>
$ AGENT=suspec-reviewer.md
$ DEST="$REPO"/.claude/agents
$ mkdir -p "$DEST"
$ cp agents/"$AGENT" "$DEST"/
# optional hooks
$ DEST="$REPO"/.claude/hooks
$ mkdir -p "$DEST"
$ cp hooks/*.sh "$DEST"/
These are Claude Code agents in .claude/agents/. npx skills installs the skills catalog, not these.
tier-1.conf — read-only workers
Read-only workers
These drop Edit/Write from the tool list. The guard narrows shell writes. It is not a sandbox.
review requiredtoolable limits
scope.note — toolable, not a wall
Limits
Tool lists and hooks narrow the surface. They do not make a hard security boundary.
delegation trace
Why not just the built-in reviewer?
Use built-ins when they are enough. Use these when a role file or delegation trace helps.
portability
Claude Code first
suspec agents emit --codex generates Codex agent files from the same definitions.
Tool scoping and hooks are specific to Claude Code. Other runners need their own controls.
catalog — browse on github
The full catalog
Six workers and two hooks. Read the file before you copy it.