evidence-anchors

evidence-anchors is a skill for Claude Code, Codex from armanfatemi/nullius. It costs 104 tokens per session (2,350 once invoked), scanned A, original, MIT.

A writing guide for adding checkable evidence to documents that describe an existing codebase. An evidence anchor points a claim to the file and location that supports it.

In plain words
What is it for?
Use it when writing design documents, RFCs, ADRs, risk assessments, plans, or pull-request descriptions that make claims about existing code.
Why use it?
It helps reviewers verify that plans, proposals, and other documents are based on what the code actually does.

Skill for Claude CodeCodex

Written for no agent in particular: nothing here depends on one.

Part of the nullius plugin — 2 skills, 3 commands, 7 hooks shipped together

Good fit Use it when writing design documents, RFCs, ADRs, risk assessments, plans, or pull-request descriptions that make claims about existing code.

Compare 6 skills from other repositories ↓
Install with agentmods
npx agentmods add skills/armanfatemi/nullius/evidence-anchors
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.

Any agent
npx skills add armanfatemi/nullius --skill evidence-anchors
Clone the repo
git clone --depth 1 https://github.com/armanfatemi/nullius

Made for: Claude Code, Codex.

Or install nullius, the plugin that ships this one along with the rest of its 2 skills, 3 commands, 7 hooks.

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

agentmods badge for evidence-anchors

README.md
[![agentmods](https://agentmods.dev/badge/skills/armanfatemi/nullius/evidence-anchors.svg)](https://agentmods.dev/skills/armanfatemi/nullius/evidence-anchors)
Your own site
<a href="https://agentmods.dev/skills/armanfatemi/nullius/evidence-anchors"><img src="https://agentmods.dev/badge/skills/armanfatemi/nullius/evidence-anchors.svg" alt="Measured on agentmods" height="20"></a>
Per session 104 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 2,350 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. A grade says what 26 rules found in the file — not that it is safe.
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 5.1 $0.00104 $0.02350
Opus 5 $0.00052 $0.01175
Sonnet 5 $0.00021 $0.00470
Haiku 4.5 $0.00010 $0.00235

Measured today against content hash 0f1c8d161b1a, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-07, from the pricing page.

Security

Grade A, and why

evidence-anchors 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 today.

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.

plugin/skills/evidence-anchors/SKILL.md · 197 lines

How it starts

The opening of the file, as written. The whole thing — 197 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Evidence Anchors — grounding claims about existing code

You are writing a document that asserts things about the codebase. Reviewers — human or agent — will check whether your plan is good, but almost nobody re-checks whether your premises are true. A false premise that supports a correct-looking conclusion sails through every review. This skill closes that hole: every load-bearing claim carries a citation that forces you to open the file, and a deterministic checker re-verifies the citation afterward.

The convention attaches to anything a human approves, not just formal design docs. That includes a plan written in plan mode — ephemeral is fine; the anchor gates the approval moment, not the archive — and a PR description, which is the one claim-carrying document every workflow has. When a plan or PR body asserts something load-bearing about existing code, anchor it.

Full spec: https://github.com/armanfatemi/nullius/blob/main/spec/evidence-anchors.md

What needs an anchor

A load-bearing claim is a statement about what the existing code, config, or infrastructure does, on which a design decision rests — anything in a rationale, an "alternatives considered … rejected because …", a risk assessment, or a constraint.

  • "The deployment runs 2 replicas" → load-bearing, cite it.
  • "PaymentProcessor already owns a retry queue" → load-bearing, cite it.
  • "No service reads this collection" → load-bearing absence, cite the search.
  • "Option B is simpler to reason about" → judgment, NO anchor.
  • "Users may find this confusing" → product risk, NO anchor.

The rule: if you cannot cite it, you may not assert it. Move the claim to an ## Open questions section instead. Do not decorate judgment calls with citations — citation theater trains readers to skim past the anchors.

The presence form — the thing exists

Open the file. Then write:

**Evidence:** `path/to/file.ext:LINE` — `exact text appearing on that line`
  • Path is repo-relative (never absolute, never ~, never ..).
  • The quoted text must actually appear on that line (whitespace is normalized, so indentation differences are fine).
  • If the cited text itself contains a backtick, use a double-backtick span: **Evidence:** `libs/x.ts:4` —const q = query {};
  • Quote something that could be wrong. Matching is substring-based, so a one- or two-character quote is trivially true and asserts nothing; it verifies as WEAK-ANCHOR. Quote enough of the line that a real change to the code would contradict it.
  • Quote something that occurs once. A quote matching several lines is WEAK-ANCHOR while it sits on its cited line, and a hard UNPINNED failure once that line number goes stale — at that point neither half of the citation identifies anything. A stale line number on its own is not a failure: DRIFT and WRONG-LINE pass, and tell you the citation needs updating.
  • Stamp the commit you read. Run git rev-parse --short HEAD and put it in the anchor:

Read the full file on GitHub · 197 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. today Changed · +2 tokens per session 0f1c8d161b1a
  2. 7d ago First seen · 197 lines · 102 tokens per session scan A 0263b895b707

Subscribe to this mod's changes

evidence-anchors is a skill published in the GitHub repository armanfatemi/nullius (5 stars, last pushed today), licensed MIT. It adds 104 tokens to every session and 2,350 once invoked, about $0.0005 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.

Related

Other skills, from other repositories

argot-setup

Set argot up for a repository end to end — audit its history, decide what should shape its voice, fit, verify the fit actually catches things, tune the rules its own history says are noisy, and wire the places it runs (pre-write hook, pre-commit, MCP, CI). One sitting, one decision at a time, each proposed with the…

get-tmonier/argot · 129 tokens

argot-check

Score your working changes with argot — flag code foreign to this repo's own patterns (unfamiliar dependencies, APIs, constructs), functions the repo already has, code filed in the wrong place, imports that break the repo's layering, and tests weakened, disabled, or deleted alongside a production change — before…

get-tmonier/argot · 110 tokens

argot-write-rule

Codify a repo convention argot's built-ins don't cover into a scripted custom rule — a .argot/rules/NAME/rule.toml manifest plus a sandboxed Rhai script that fires exactly like a built-in, gated on a green fixture suite before it ever sees a real diff. Use when the user asks to "write an argot rule for X", "codify…

get-tmonier/argot · 131 tokens

argot-review-pr

Review a specific pull request (or diff range) against this repo's learned patterns with argot, without checking it out — flag dependencies, APIs, and constructs foreign to how the repo is written, duplicated functions, misfiled code, layering breaks, and tests weakened, disabled, or deleted alongside a production…

get-tmonier/argot · 78 tokens

argot-suggest-rules

Turn a convention argot has already discovered into a scripted custom rule. Runs argot conventions, surfaces the repo's mined placement conventions (where a kind of code lives — "validation in schema files", "DB access only in migrations", "business logic in the service layer, not views") and internal-API vocabulary…

get-tmonier/argot · 173 tokens

bellwether

Bring the current PR to a mergeable state: CI green, all review comments resolved, no merge conflicts. Self-contained — watches CI, fixes issues, watches again until merge-ready. Use when the user wants to keep a PR green, auto-fix CI, resolve review comments, or says "get this merged".

roderik/bellwether · 68 tokens