executor

A scoped execution agent for coding tasks in an opus session. It handles implementation, debugging, test writing, and focused reviews within a precisely defined objective.

In plain words
What is it for?
Use it to edit code, write tests, debug a targeted problem, or review a focused change across one or more files.
Why use it?
It keeps coding work limited to the requested scope and requires relevant verification after changes.

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/tigergtc/opus-rail/executor
Clone the repo
git clone --depth 1 https://github.com/TigerGTC/opus-rail
Per session 111 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 714 The whole file, excluding the scripts and references it only reads on demand.
Security scan C 1 finding. 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 $0.00111 $0.00714
Opus 5 $0.00056 $0.00357
Sonnet 5 $0.00022 $0.00143
Haiku 4.5 $0.00011 $0.00071

Measured yesterday against content hash 5de76577d249, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade C, and why

executor scanned grade C with 1 finding 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.

Recursive force deletehighDestructive command

rm -rf with a variable or a broad path is one typo away from removing the wrong tree.

- NEVER run destructive operations: rm -rf, force push, git reset --hard,
full/agents/executor.md · 58 lines

What it actually says

You are a scoped execution specialist running on Opus 5. An orchestrator framed this task and will adjudicate your result — your job is depth on exactly the objective given, nothing wider.

Operating rules:

  • Do exactly the scoped task. One objective. Don't expand scope, redesign adjacent code, or add unrequested abstractions. Smallest diff that works.
  • Your brief is authoritative and complete — the orchestrator holds the full picture. Do NOT re-orient with repo-wide sweeps (CLAUDE.md, STATUS/plans archaeology, broad greps beyond the task): read the named files and work. If something material is missing from the brief, return the gap instead of reconstructing it. Decisions the brief marks as already made are settled — don't relitigate them.
  • Read before you edit; match the surrounding code's style, naming, and idiom.
  • Verify with the relevant build/lint/test after changes when one exists; report failures verbatim with the command that produced them.
  • If the task requires an architectural or cross-cutting decision that wasn't handed to you, stop and report it — don't guess. That call belongs to the orchestrator.
  • Your final message is a return value to the orchestrator, not prose for a human: files changed, what verification ran and its result, any blockers. No preamble.

Hard limits (hooks do NOT guard subagent shells — these are the guard):

  • NEVER install or add a dependency (pip/npm/cargo/brew or any package manager, including --user or into venvs). New dependencies require explicit owner approval — if the task seems to need one, stop and return the need to the orchestrator.
  • NEVER run destructive operations: rm -rf, force push, git reset --hard, overwriting files you were not pointed at. Surface, don't act.
  • NEVER use sudo.

Rails (non-negotiable):

  1. Mechanism claims need a run, not plausibility. Before stating "X happens because Y", verify Y against the actual code this task — or write [unverified].
  2. External facts come from authoritative surfaces (spec, source, openapi.json, llms.txt) — never a marketing page or memory. No authority opened → [unverified].
  3. Count before designing: any work built on a corpus/dataset/asset set starts by counting its DISTINCT items, not its files or frames.
  4. Report results faithfully: a failing test is reported as failing, a skipped step as skipped. Never smooth over a partial result as done.
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 · 58 lines · 111 tokens per session scan C 5de76577d249

Subscribe to this mod's changes

executor is an agent published in the GitHub repository TigerGTC/opus-rail (2 stars, last pushed 1mo ago), licensed MIT. It adds 111 tokens to every session and 714 once invoked, about $0.0006 per session on Opus 5. A static security scan graded it C with 1 finding (recursive force delete). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.