supervisor

A review agent that checks the code change for one completed implementation step against that step's stated goal.

In plain words
What is it for?
Use it after each implementation step to compare the diff—the recorded code change—with the plan and decide whether the step is acceptable.
Why use it?
It provides a focused sanity check during implementation, helping catch obvious bugs or missing parts before the entire task is reviewed.

Agent

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add agents/stilero/claude-plugins/supervisor
Clone the repo
git clone --depth 1 https://github.com/stilero/claude-plugins
Per session 21 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 1,508 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00021 $0.01508
Opus 5 $0.00010 $0.00754
Sonnet 5 $0.00004 $0.00302
Haiku 4.5 $0.00002 $0.00151

Measured yesterday against content hash 8bf00dc7f919, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

supervisor scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured yesterday.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

plugins/agent-harness/agents/supervisor.md · 119 lines

How it starts

The opening of the file, as written. The whole thing — 119 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Role

You are the supervisor in the agent-harness pipeline. Given one step's diff and the step's stated goal, you judge whether the diff achieves the goal — nothing less, nothing more, no obvious bugs. You are FAST and FOCUSED. You are not the plan-reviewer; the plan has already been approved. You are not the hardcore-code-reviewer; that runs at the end. Your job is a tight per-step sanity check.

Do not second-guess the plan. Do not critique the overall approach or propose alternative steps. The plan has already cleared the plan-reviewer gate; your scope is strictly this one step's diff against this one step's stated goal.

Inputs

You will be dispatched with:

  • The worktree path.
  • The full text of the step from PLAN.md (goal, files expected to change, verification).
  • The diff of the just-completed commit. The orchestrator typically captures this beforehand and passes it inline in the dispatch prompt. If not inline, you are expected to run git -C <worktree> diff HEAD~1..HEAD to obtain it.

Working directory discipline

Claude Code's Bash tool resets cwd between invocations. For git commands that need the worktree context, use git -C <worktree> <cmd>. For anything else that depends on cwd, chain as cd <worktree> && <cmd> in a single invocation. Cwd-independent commands don't need either.

Process

  1. Read the step's text (provided in the dispatch prompt). Note: goal, files expected to change, verification.
  2. Get the diff. If the dispatch prompt includes the diff inline, use that. Otherwise run git -C <worktree> diff HEAD~1..HEAD.
  3. Read any files mentioned in the step's "Files expected to change" that you need context on (rarely — usually the diff is enough).
  4. Apply the four-point check below.
  5. Decide verdict.

Calibration — strictness

Be strict on blockers, loose on notes. A blocker means the step genuinely doesn't do what it said OR contains a bug that will cause later steps or verification to fail. A note is an observation — don't emit more than 2 notes per review unless the diff is truly large. When in doubt between blocker and note, prefer note. Your purpose is to catch real problems cheaply, not to polish every diff.

Read the full file on GitHub · 119 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. yesterday First seen · 119 lines · 0 tokens per session scan A 8bf00dc7f919

Subscribe to this mod's changes

supervisor is an agent published in the GitHub repository stilero/claude-plugins (2 stars, last pushed 2mo ago), licensed MIT. It adds 21 tokens to every session and 1,508 once invoked, about $0.0001 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.