conviction

A structured review of a belief, claim, or decision. It lists the assumptions and evidence behind it, estimates confidence, and defines a test that could show the belief is wrong.

In plain words
What is it for?
Use it to audit product choices, technical plans, business claims, or other decisions where you need to know what supports them and what could disprove them.
Why use it?
It turns an unclear opinion into something you can examine and test. This helps reveal weak evidence and unspoken assumptions before you commit to a decision.

Command

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 commands/nicoladevera/thinking-stack/conviction
Clone the repo
git clone --depth 1 https://github.com/nicoladevera/thinking-stack
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 1,654 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 $0.01654
Opus 5 $0.00010 $0.00827
Sonnet 5 $0.00004 $0.00331
Haiku 4.5 $0.00002 $0.00165

Measured yesterday against content hash 9060db86cd2f, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

conviction 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.

conviction/commands/conviction.md · 178 lines

How it starts

The opening of the file, as written. The whole thing — 178 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Conviction — Epistemic Belief Audit

You are running a conviction audit: a structured epistemic analysis that takes a belief, claim, or decision in $ARGUMENTS and turns it into a falsifiable scorecard — mapping assumptions, weighing evidence, scoring confidence, and designing the cheapest test that could prove you wrong.

The user invoked this with: $ARGUMENTS


Phase 1: Intake & Readiness

Assess whether $ARGUMENTS contains a clear belief or claim. You need: (1) a stated belief, claim, or decision, and (2) enough context to reason about what would make it true or false.

If the input is too thin (no discernible claim, purely emotional, no domain context): Use AskUserQuestion to ask up to 3 targeted clarifying questions. Only ask what's actually missing:

  1. What is the belief or claim you're trying to evaluate?
  2. What domain or context does this apply to? (e.g., your team, a product, a market, a personal decision)
  3. What are the stakes or time window, if relevant? (low/medium/high — helps calibrate how much rigor to apply)

If the input is ready: proceed directly.

Once you have enough context, restate the claim in falsifiable form — a precise statement that could, in principle, be shown to be true or false. Surface it before proceeding (no confirmation required — just present it and move on):

Claim (falsifiable): [Precise, falsifiable restatement of the belief]


Phase 2: Research (Optional, Targeted)

Determine whether external data would meaningfully ground the analysis: base rates, analogous comparisons, published benchmarks, or documented evidence relevant to the claim.

If yes: run 2–3 targeted WebSearch / WebFetch queries. Surface relevant findings in a brief paragraph before the audit. Focus on:

  • Empirical data or benchmarks directly bearing on the claim
  • Analogous cases where similar beliefs were tested and what the outcome was
  • Known base rates for the domain (e.g., adoption rates, failure rates, performance benchmarks)

Read the full file on GitHub · 178 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 · 178 lines · 21 tokens per session scan A 9060db86cd2f

Subscribe to this mod's changes

conviction is a command published in the GitHub repository nicoladevera/thinking-stack (2 stars, last pushed 4mo ago), licensed MIT. It adds 21 tokens to every session and 1,654 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.