A command-line tool for checking whether claims in documents are supported by source files. It can check one claim, one document, or groups of documents described in a YAML map.

Use it to extract claims, compare them with source documents, check consistency, and produce grounding reports for one or many documents.

It reduces the risk of presenting unsupported or inconsistent statements as facts. It also creates reports showing how claims relate to their sources.

grounding — Command by stellarshenson

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/stellarshenson/claude-code-plugins/grounding

Clone the repo

git clone --depth 1 https://github.com/stellarshenson/claude-code-plugins

Per session 22 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.

When invoked 1,202 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.

Model	Per session	Once invoked
Fable 5	$0.00022	$0.01202
Opus 5	$0.00011	$0.00601
Sonnet 5	$0.00004	$0.00240
Haiku 4.5	$0.00002	$0.00120

Measured 2d ago against content hash bd1a7aa06ba8, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

grounding 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.

plugins/document-processing/commands/grounding.md · 71 lines

How it starts

The opening of the file, as written. The whole thing — 71 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Grounding

Invoke document-processing:grounding skill. Pure grounding - no tone/style/format compliance (that = /document-processing:validate). Skill always runs document-processing CLI; generative interpretation only on-top layer for semantic claims.

Three modes (skill picks from argument):

Single claim - document-processing ground --claim "..." --source <file> --json
One document — full grounding chain (extract → ground → consistency) - document-processing extract-claims --document <doc> --output validation/claims.json -> review claims.json -> document-processing ground --manifest validation/claims.json --source <src1> --source <src2> --output validation/grounding-report.md -> document-processing check-consistency --document <doc> --output validation/consistency-report.md. All three steps run; run produces both grounding-report.md and consistency-report.md.
Many documents - source_map.yaml declaring clients[].sources + document (+ optional primary_source) -> document-processing validate --manifest source_map.yaml --output-dir validation/, runs same chain per client, producing validation/<client>/{claims.json,grounding-report.md,consistency-report.md}

Default engine

Lexical mode is the default: a frozen-weight logistic over 13-18 signals selected by lexical_effort (low / medium / high, default high). All lexical tiers are CPU-only with no extra install required. Use --effort CLI overlay or set lexical_effort in config to switch tiers.

Optional layers (opt-in)

semantic - --semantic adds bge-m3 retrieval + reranker (OpenVINO int8, torch-free)
NLI / entailment - the truth signal: cross-encoder reads (evidence, claim) -> grounded / unconfirmed / contradicted; multilingual, catches contradictions + cross-lingual matches lexical misses
calibration - tune the verdict to your corpus from labelled evidence: calibrate --action update --evidence f.json then config set-calibrator; learned weights live in config. Public-data check: make grounding-validate ENGINE=nli. Full doc: docs/grounding_calibration.md

Read the full file on GitHub · 71 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.

2d ago First seen · 71 lines · 22 tokens per session scan A bd1a7aa06ba8

Subscribe to this mod's changes

grounding is a command published in the GitHub repository stellarshenson/claude-code-plugins (3 stars, last pushed 3d ago), licensed MIT. It adds 22 tokens to every session and 1,202 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.

Research

grounding

Getting it into your agent

What it costs to keep this loaded

Grade A, and why

Nothing flagged

How it starts

Grounding

Default engine

Optional layers (opt-in)

What this file has done since we first saw it

Other commands, from other repositories

git

checklist

clarify

specify

analyze

converge