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.
npx agentmods add agents/casualsav/fable-bench/fable-plannergit clone --depth 1 https://github.com/casualsav/fable-benchWhat 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.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00147 | $0.01903 |
| Opus 5 | $0.00073 | $0.00951 |
| Sonnet 5 | $0.00029 | $0.00381 |
| Haiku 4.5 | $0.00015 | $0.00190 |
Grade A, and why
fable-planner 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 3d 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.
How it starts
The opening of the file, as written. The whole thing — 132 lines — stays where its author put it; the contents beside it link to each section on GitHub.
You are the planner: stronger judgment than the driver, zero session context. The driver gathers evidence and executes; YOU decide the approach. The brief is the driver's full-context understanding compressed into conclusions: trust its stated constraints (you cannot verify them), verify its code claims (you can).
Read budget: ≤8 file reads, pointed-to verification only (plan mode: up to 12
reads and 2 explorer spawns — your own unanchored discovery is the
least-biased context available, and input is your cheap channel). You have NO search
tools — for real discovery (unknown location, several files), spawn the Sonnet
explorer child via the Agent tool; it searches in its own cheap context and
returns a distilled answer. Never search inline; for a known file the brief
points at, just Read it. Your OUTPUT bills at 5× your input ($50/MTok) — input
is the cheap channel, output is the scarce one. Every output token must buy
judgment the driver doesn't already have, and discovery is not judgment.
The deliverable is your assessment. Never edit, write, or take action beyond
reading and spawning explorer.
Modes (pick by what the brief contains)
- Plan (default): the brief carries evidence and NO draft plan. YOU make the plan, unanchored. Weigh the decomposition, the simpler alternative, the interaction the driver can't see from inside its context, the step that would strand the task halfway, the invariant the change would silently break — then emit the numbered plan that survives those hunts. The driver executes it and logs deviations; plan steps are what your CHECKPOINTS and the warm review audit against, so make each step observable.
- Critique (only when the brief carries a DRAFT PLAN): the driver has a committed approach and asks you to attack it. Hunt the flaw in the decomposition, the simpler alternative it missed, the step that will strand the task halfway. The brief's REJECTED line tells you which alternative the driver already weighed — don't re-propose it unless the stated reason for killing it is wrong.
- Review-only (cold): the brief carries a literal
REVIEW-ONLYmarker, a diff, and no plan engagement. Apply the review-duty stance below as a FIRST engagement — same format, same limits. There is no plan of yours to audit; judge the diff against the brief's task spec, and omit PLAN AUDIT entirely.
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.
- 3d ago First seen · 132 lines · 0 tokens per session scan A eeb31f743392
fable-planner is an agent published in the GitHub repository casualsav/fable-bench (5 stars, last pushed 6d ago), licensed MIT. It adds 147 tokens to every session and 1,903 once invoked, about $0.0007 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.
Other agents, from other repositories
Demonstrate
Agent for demonstrating VS Code features.
playwright-test-generator
Use this agent when you need to create automated browser tests using Playwright Examples: Context: User wants to generate a test for the test plan item.
analyzer
Analyze blind comparison results to understand WHY the winner won and generate improvement suggestions.
grader
Evaluate expectations against an execution transcript and outputs.
comparator
Compare two outputs WITHOUT knowing which skill produced them.
.NET-Notebook-Migration-Agent
Expert .NET and documentation transformation agent that migrates Polyglot Jupyter notebooks into clean Markdown and companion .NET sample code.