dedupe-analyze

dedupe-analyze is a command for coding agents from axiomantic/spellbook. It costs 43 tokens per session (3,871 once invoked), scanned A, original, MIT.

An analysis command for finding duplicate or closely related Markdown blocks. It filters possible pairs in stages and sends each remaining pair to a classifier for a verdict.

In plain words
What is it for?
Run it after setup to classify candidate duplicate blocks, before generating the review report.
Why use it?
It reduces the number of comparisons needed and records cases that must be kept because of safety or document-structure rules.

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/axiomantic/spellbook/dedupe-analyze
Clone the repo
git clone --depth 1 https://github.com/axiomantic/spellbook

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 dedupe-analyze

README.md
[![agentmods](https://agentmods.dev/badge/commands/axiomantic/spellbook/dedupe-analyze.svg)](https://agentmods.dev/commands/axiomantic/spellbook/dedupe-analyze)
Your own site
<a href="https://agentmods.dev/commands/axiomantic/spellbook/dedupe-analyze"><img src="https://agentmods.dev/badge/commands/axiomantic/spellbook/dedupe-analyze.svg" alt="Measured on agentmods" height="20"></a>
Per session 43 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 3,871 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 5.1 $0.00043 $0.03871
Opus 5 $0.00022 $0.01936
Sonnet 5 $0.00009 $0.00774
Haiku 4.5 $0.00004 $0.00387

Measured 5d ago against content hash 3e4772f44e8a, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-05, from the pricing page.

Security

Grade A, and why

dedupe-analyze 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 5d 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.

commands/dedupe-analyze.md · 411 lines

How it starts

The opening of the file, as written. The whole thing — 411 lines — stays where its author put it; the contents beside it link to each section on GitHub.

MISSION

Phase 2 of the dedupe skill: consume the blocks manifest produced by /dedupe-setup, narrow the O(N²) pair space through a staged pipeline, and dispatch one classifier subagent per surviving pair to produce a verdict file.

Part of the dedupe- command family.* Run after /dedupe-setup. Run before /dedupe-report.

Invariant Principles

  1. Narrowing is staged — each pair that reaches the classifier has passed every prior gate in order.
  2. Mechanical safety floor short-circuits classification — any pair with a block flagged inline_mandatory_mechanical=true is recorded as a KEEP-flavored verdict with source=mechanical_floor without a classifier dispatch.
  3. Structural template floor short-circuits classification — any intra-bucket pair whose heading-topic bucket key (derived per skills/dedupe/references/segmentation-protocol.md §3) matches the allowlist in skills/dedupe/references/template-headings.md is recorded as KEEP-placement with source=structural_template without a classifier dispatch. Cross-bucket pairs surfaced by triage are exempt to preserve drift detection.
  4. One subagent per surviving pair — each pair gets exactly one Task dispatch with a per-pair randomized 16-hex-char sentinel nonce wrapping the inert-DATA blocks.
  5. Off-schema fails safe toward KEEP — verdict-scoped only; the coercion rule is defined in the canonical home (see References).
  6. Triage off-schema HALTS — a malformed triage response means the cross-bucket signal is unreliable; halting is the safety-preserving choice. There is no triage coercion.
  7. Off-schema halt threshold is 25% — running rate over the last 20 verdicts (or all verdicts if M < 20). Exceeding the threshold HALTS analyze with a failure report. This is NOT verdict coercion; the halt produces a failure artifact, not a misleading clean run.

Halt-not-coerce invariant: The 25% off-schema halt threshold and the triage-off-schema HALT are correctness-preserving safety stops. Producing a coerced-clean report when the cross-bucket signal or the per-pair classifier signal is degrading would mask exactly the failures the skill exists to surface. HALT is the correct response, not best-effort completion.

Read the full file on GitHub · 411 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. 5d ago First seen · 411 lines · 43 tokens per session scan A 3e4772f44e8a

Subscribe to this mod's changes

dedupe-analyze is a command published in the GitHub repository axiomantic/spellbook (10 stars, last pushed yesterday), licensed MIT. It adds 43 tokens to every session and 3,871 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 2026-08-31.