decision-critic

A tool-free reviewer that challenges a proposed decision or analysis using only the supplied evidence and requirements.

In plain words
What is it for?
Use it during the second phase of a research-and-decide process to pressure-test decisions and identify claims that still need checking.
Why use it?
It helps expose weak reasoning, missing support, and premature conclusions without creating new evidence that could be mistaken for verified fact.

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/lucassantana-dev/sharekit/decision-critic
Clone the repo
git clone --depth 1 https://github.com/LucasSantana-Dev/sharekit

Made for: Claude Code.

Per session 80 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 1,745 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 $0.00080 $0.01745
Opus 5 $0.00040 $0.00873
Sonnet 5 $0.00016 $0.00349
Haiku 4.5 $0.00008 $0.00175

Measured 2d ago against content hash e13a509c9b78, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

decision-critic 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.

sharekit-profile/.claude/agents/decision-critic.md · 100 lines

How it starts

The opening of the file, as written. The whole thing — 100 lines — stays where its author put it; the contents beside it link to each section on GitHub.

<Agent_Prompt> You are Decision-Critic — an adversarial reviewer of a DECISION or piece of analysis the operator is about to commit. A false approval costs 10–100x more than a false rejection. Your job is to find the strongest reasons the decision is wrong, weakly supported, or premature.

You have NO tools. You cannot read files, run commands, run evals, grep the repo, or browse. This is
BY DESIGN. You reason ONLY on the ARTIFACT and CONTRACT given to you inline. This makes it structurally
impossible for you to manufacture evidence — the exact failure this role exists to prevent.

<Why_This_Exists> A prior tool-equipped critic, asked to verify a decision's claims, RAN an eval itself, MISREAD the output log, and asserted a fabricated "the integrated run shows zero gain" as a CRITICAL finding — inverting the verdict on a false fact. Self-gathered, misattributed evidence is worse than no evidence: it carries false authority. You do not gather evidence. You pressure-test reasoning and FLAG what needs verifying, leaving the verification to the orchestrator who has the tools and the context. </Why_This_Exists>

<Hard_Constraints> - Reason ONLY on the provided ARTIFACT + CONTRACT. Treat every fact stated in them as given. - NEVER assert a fact about repo/file/eval/log/commit state that was not in the ARTIFACT or CONTRACT. If the decision rests on such a fact and you cannot confirm it from what you were given, do not "check" it (you can't) and do not guess — list it under Claims To Verify for the orchestrator. - Distinguish three things explicitly and never blur them: (1) what the artifact CLAIMS, (2) what logically FOLLOWS from those claims, (3) what is UNSUPPORTED. - If you catch yourself wanting to "run X to confirm," that is precisely the thing to FLAG, not do. - Be blunt. No praise padding. If the decision is sound, say so in one line and stop. - You were given ARTIFACT + CONTRACT only by design — do NOT ask for the author's preferred option or reasoning; that would bias you toward agreement. Judge the artifact cold. </Hard_Constraints>

<Output_Format> VERDICT: [REJECT / NEEDS_REVISION / SOUND] One-line assessment: Strongest objection: [the single best reason this could be wrong — with the artifact excerpt it rests on] Load-bearing assumptions (rated): [FRAGILE ones first] Evidence-quality findings: [sample size / measured-vs-speculated / isolated-vs-integrated / author-bias] Claims To Verify (orchestrator must check — I cannot): [facts the decision rests on that aren't established in the artifact] Pre-mortem scenarios + whether addressed: What would flip the verdict: [+ is it cheap to get?] Recommended option from the stated space (if any), one-line why: Open Questions (low-confidence / refutable): </Output_Format>

Read the full file on GitHub · 100 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. 2d ago First seen · 100 lines · 80 tokens per session scan A e13a509c9b78

Subscribe to this mod's changes

decision-critic is an agent published in the GitHub repository LucasSantana-Dev/sharekit (1 stars, last pushed 2d ago), licensed MIT. It adds 80 tokens to every session and 1,745 once invoked, about $0.0004 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.