headwater-sweep

Audit a documentation corpus slice for contradictions, supersessions, and undefined concepts.

Updated Aug 8, 2026
One-click install
npx skills add https://github.com/headwater-ai/headwater --skill headwater-sweep-headwater-ai
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: headwater-sweep
Source: https://github.com/headwater-ai/headwater/tree/main/.claude/skills/headwater-sweep
Command: npx skills add https://github.com/headwater-ai/headwater --skill headwater-sweep-headwater-ai

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Structural checks can verify that links resolve and identifiers bind, but they cannot tell whether a documentation corpus adds up to one coherent account. This Skill performs the human-model half of a coherence sweep: reading documents and reporting the contradictions, quiet supersessions, and gaps that no deterministic linter can detect. ## Core Features & Use Cases - Briefing-driven workflow: Run headwater sweep plan --under <path> to get every classified document in a slice with its identifier, kind, title, summary, and all declared graph edges. - Five finding classes: Report undeclared_conflict, quiet_supersession, undefined_concept, audience_mismatch, and unwritten_section findings with verbatim quotations the engine can verify. - Validated reporting: Submit findings via headwater sweep report, which confirms each quotation exists in the cited document, each path is a classified document, and no proposed edge duplicates the graph. - Use Case: After a major spec revision, sweep docs/spec to find older documents that still read as live, then propose conflicts_with edges so the engine owns the problem permanently. ## Quick Start Ask the assistant to run a coherence sweep over the docs/spec directory and report any contradictions or quietly superseded documents it finds.

Frequently Asked Questions about headwater-sweep

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I run a coherence sweep over a documentation corpus?

Run headwater sweep plan --under <path> to get a briefing of every classified document in the slice, read all the documents, then write findings as YAML and submit them with headwater sweep report <path>.

What kinds of documentation problems can a sweep find?

A sweep reports five classes: undeclared_conflict, quiet_supersession, undefined_concept, audience_mismatch, and unwritten_section. Readability complaints and any sixth class are explicitly out of scope.

Why was my sweep finding refused by the report command?

Refusals happen for four reasons: a quotation the document does not contain, a path that is not a classified document, a class outside the five allowed, or a proposal the graph already declares. Quote passages verbatim rather than paraphrasing.

Can a sweep fail a CI build or gate a release?

No. headwater sweep report exits 0 regardless of findings, and no CI job runs it. Output is a proposal a person reads and accepts, never a verdict anything gates on.

Can I sweep source code directories like the engine tree?

No. The census types only documents, so --under engine names nothing and intake refuses paths that are not classified documents. Sweeps operate on typed documentation slices only.