sd-coherence-audit

Audit a document corpus for contradictions, vagueness, bandaid guidance, and redundancy.

Updated Jun 26, 2026
One-click install
npx skills add https://github.com/platypeeps/sd-ai-command-pack --skill sd-coherence-audit-platypeeps
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: sd-coherence-audit
Source: https://github.com/platypeeps/sd-ai-command-pack/tree/main/contrib/sd-coherence-audit
Command: npx skills add https://github.com/platypeeps/sd-ai-command-pack --skill sd-coherence-audit-platypeeps

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? Knowledge corpora such as note vaults, agent-instruction files, and documentation trees accumulate internal defects over time: passages that contradict each other, directives too vague to act on, temporary workarounds fossilized into policy, and the same rule restated in several places. This Skill audits a bounded corpus against itself and returns a read-only findings ledger with both sides of every conflict quoted. ## Core Features & Use Cases - Four defect detectors: contradiction, vagueness, bandaid, and redundancy, each with named criteria and near-miss rules that prevent false positives. - Precedence-aware conflict classification: distinguishes true contradictions from conflicts that a declared authority ordering settles or that lack one entirely. - Read-only, sensitivity-controlled reporting: the corpus is never edited, quotes can be redacted or withheld under restricted sensitivity levels, and partial coverage is always disclosed. - Use Case: Point the skill at a docs tree before a major release to find instructions that cannot both be followed, gates with undefined pass conditions, and duplicated rules that have drifted apart. ## Quick Start Run sd-coherence-audit with input=docs/,CONTRIBUTING.md to audit those files for contradictions, vagueness, bandaid guidance, and redundancy and return a findings ledger.

Frequently Asked Questions about sd-coherence-audit

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

FAQPage Schema
How do I audit documentation for contradictions?▼

Run the audit with input= pointing at your docs tree or file list. The skill builds an index of directives and assertions, compares entries grouped by subject, and reports contradictions with both sides quoted at their path:line locations.

What kinds of documentation defects can be detected automatically?▼

Four classes are detected: contradictions between passages, vague directives with no actionable trigger, bandaid guidance shaped as patches rather than rules, and redundant restatements of the same rule. Each finding names the specific criterion it satisfies.

Does the audit modify my files?▼

No, the audit is strictly read-only. It never edits, reorganizes, or reformats the corpus, and proposed resolutions in the findings ledger are suggestions that are never applied.

Can I exclude files or limit which detectors run?▼

Yes. Use exclude= to remove paths or file classes from the audit boundary, and classes= to run only a subset of the contradiction, vagueness, bandaid, and redundancy detectors. Unknown argument names are reported as errors before any reading.

How are false-positive contradictions avoided?▼

Each detector has near-miss rules that drop candidates such as compatible thresholds, explicit exceptions, non-overlapping scopes, and conflicts settled by declared precedence. Findings below medium confidence are dropped and counted rather than reported.

What happens when the corpus contains secrets or sensitive text?▼

Secret-shaped values inside quotes are redacted while keeping the finding legible. Under sensitivity=restricted named sensitive material is withheld, and under sensitivity=minimal quotes are omitted entirely with the finding marked unquotable rather than dropped.