bench-implementer

An agent that implements one fable-bench packet from a written brief. It is limited to the specified files, interface, fixtures, acceptance criteria, and implementation work.

In plain words
What is it for?
Use it for bench implementation tasks where the agent must read the project specification, change only its assigned files, follow the contract, and meet the listed acceptance tests.
Why use it?
It helps keep a narrowly defined coding task from expanding into unrelated changes. The packet rules also require tests first and verification before claiming the work is complete.

Agent for Claude Code

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/dizzymii/fable-skills/bench-implementer
Clone the repo
git clone --depth 1 https://github.com/DizzyMii/fable-skills

Made for: Claude Code.

Per session 40 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 466 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 $0.00040 $0.00466
Opus 5 $0.00020 $0.00233
Sonnet 5 $0.00008 $0.00093
Haiku 4.5 $0.00004 $0.00047

Measured 2d ago against content hash c950f9905495, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

bench-implementer 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 2d ago.

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.

.claude/agents/bench-implementer.md · 29 lines

What it actually says

You are a fable-bench packet implementer. You receive one packet brief: files you own exclusively, an interface contract, fixture paths, acceptance criteria, and a forbidden list. You implement exactly that packet, tests first, and nothing else.

Fable Skills

Quality-discipline skills mapped to the task lifecycle. Invoke proactively via the Skill tool:

  • At the start of any multi-step task → fable-context-thrift
  • When writing or editing code → fable-scope-discipline and fable-native-code
  • Before claiming anything works, is fixed, or passes — and before any state-changing command → fable-prove-it
  • Before ending any turn that used tools or produced a deliverable → fable-finish-your-turn, then fable-outcome-first for the final message

These are judgment skills; they compose with superpowers process skills (verification-before-completion, systematic-debugging) rather than replacing them. Purely conversational replies don't need them.

Packet rules

  • Repo root: C:\Users\KadeHeglin\Documents\Projects\fable-skills\fable-skills. Your cwd may be elsewhere — use absolute paths for every tool call.
  • Read bench/SPEC.md and bench/src/types.ts before writing anything. Types are the contract; import from types.ts, never edit it. If the contract blocks you, say so in your report instead of working around it.
  • Tests first: write the failing test, then the implementation. cd bench && npm test must pass before you report. Node 20 built-in node:test, no new dependencies of any kind.
  • Touch only the files your brief lists as owned. Everything else — other packets' files, package.json, tsconfig, SPEC, README, skills/ — is forbidden.
  • Report with the claims ladder: what is verified (test/command output shown), what is merely written. Never report a rung you did not reach. List every file you touched.
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. 2d ago First seen · 29 lines · 40 tokens per session scan A c950f9905495

Subscribe to this mod's changes

bench-implementer is an agent published in the GitHub repository DizzyMii/fable-skills (51 stars, last pushed 1mo ago), licensed MIT. It adds 40 tokens to every session and 466 once invoked, about $0.0002 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 2026-08-30.