Universal document consistency auditor. Runs deterministic drift checks across
docs, code, tests, and CI, resolves conflicts through a per-fact-class authority
registry (sot-map.yaml), and produces reports that structurally separate
deterministic findings from advisory LLM-judged findings.
Use when checking documentation accuracy before a release, after major refactors,
or when onboarding to a new codebase. Also invoked automatically by /release Phase
1.6.
Trigger with "/validate-consistency", "check consistency", "validate docs",
"audit documentation", "doc drift check".
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
A direct command skips the review prompt. Inspect the source before running it.
Universal document consistency auditor. Runs deterministic drift checks across
docs, code, tests, and CI, resolves conflicts through a per-fact-class authority
registry (sot-map.yaml), and produces reports that structurally separate
deterministic findings from advisory LLM-judged findings.
Use when checking documentation accuracy before a release, after major refactors,
or when onboarding to a new codebase. Also invoked automatically by /release Phase
1.6.
Trigger with "/validate-consistency", "check consistency", "validate docs",
"audit documentation", "doc drift check".
Runs deterministic drift checks across all documentation artifacts, resolves conflicts through a per-fact-class authority registry (sot-map.yaml), and produces a structured report with actionable findings grouped by category and severity. Deterministic checks and LLM-judged checks are structurally separated: judged findings are advisory-only and never block anything.
Truth Invariant
The governed brain arbitrates asserted company/doctrine claims and the rules of arbitration โ never generated facts. This validator never reads the brain as ground truth for generated facts: authority for a generated fact lives only in a registry row a human has adjudicated. The brain ingests only human-adjudicated findings โ raw findings from this skill are candidate evidence, not truth, until a human adjudicates them.
Examples
Basic audit:
/validate-consistency
Runs the full audit on the current working directory. Loads the authority registry and produces a categorized report.
Targeted check after refactor:
I just renamed a bunch of modules โ check consistency
Detects that code changed, runs all 9 checks, surfaces broken cross-references (deterministic) and stale capability claims (advisory).
Pre-release gate (via /release):
/release
Phase 1.6 automatically invokes this skill. Deterministic Critical findings block the release; advisory findings are surfaced for review and never block.
Prerequisites
Must be run from inside a git repository (or directory with recognizable project structure)
Read access to all project files
An authority registry (sot-map.yaml) โ optional; without one the skill runs in bootstrap mode (see Step 1)
The source of truth is a per-fact-class authority registry declared as data โ never a global ranking, never auto-detected at runtime.
Resolve the registry path: the path supplied by the invoker (argument or explicit instruction) if given; otherwise the default ~/000-projects/intent-os/sot-map.yaml.
Read and parse the registry. Print what was loaded:
echo "Authority registry: [path]"
echo "Fact classes registered: [n] โ [list of class names]"
If the registry is missing or unparseable, announce bootstrap mode: every fact class is unowned. Conflicts are still detected and reported, but no winner is ever named โ each is emitted as unowned fact-class โ human adjudication needed, and the report includes drafted registry rows for a human to adjudicate (see Bootstrap).
Registry row shape:
version:1fact_classes:version:# key = fact_class (snake_case / dotted)owner:package-manifest# required โ the ONE producing surfacemirrors:# surfaces expected to restate the fact;-readme# each diffed against the owner only-changelog# (star topology โ never pairwise NรN)determinism:deterministic# deterministic | llm-judgeddepth_tier:T1# T1 value-equality | T2 judged/semanticstaleness_bound:0# days a mirror may lag before lag = driftcriticality:high# critical | high | medium (severity ceiling)volatile:false# true = mirrors expected to trail (Info-level# replication lag inside the bound)adjudicated_by:jeremy# optional provenance โ who made the calladjudicated_on:YYYY-MM-DD# optional provenance โ whenlicense:owner:license-filemirrors: [readme]
determinism:deterministicdepth_tier:T1staleness_bound:0criticality:highvolatile:false
Resolution rule: when two artifacts disagree on a fact of class F โ if the registry has a row for F, the artifact belonging to that row's owner class is correct and the finding is filed against the other artifact. If the registry has no row for F, emit the finding as unowned fact-class โ human adjudication needed: list both values and locations, name no winner, never guess.
Reference: See references/sot-registry.md for the full registry specification, fact-class and artifact-class vocabularies, and the bootstrap appendix.
Step 2: Inventory Documentation Artifacts
Scan for all documentation artifacts that will be audited:
# Find all documentation filesecho"Scanning for documentation artifacts..."
Execute ALL applicable checks from the list below. Each check is independent โ run every check that has the required files present. Skip checks whose input files don't exist (note the skip, don't error).
Every check carries a Lane tag:
Deterministic โ existence, string-equality, and exact-comparison checks. Findings may be Critical, Warning, or Info.
LLM-judged (advisory) โ checks requiring semantic judgment. Findings are โช Advisory, always โ never Critical, never blocking.
Reference: See references/drift-categories.md for category definitions and severity guide.
Check 3.1: Index vs Filesystem (Category 7 โ Index/Reference Drift)
Lane: Deterministic (referential integrity โ no registry row needed; the filesystem is the definitional referent).
If 000-docs/000-INDEX.md exists:
Read the index file
Extract all file paths listed in the index
For each path, verify the file exists on disk using Glob
Scan the 000-docs/ directory for files NOT listed in the index
Report: missing files (listed but don't exist), unlisted files (exist but not indexed)
Check 3.2: Version String Consistency (Category 1 โ Status Drift)
Lane: Deterministic. Fact class:version-string.
Collect version strings from ALL available sources:
VERSION or version.txt file
package.json โ version field
*.gemspec โ version attribute
pyproject.toml โ [project] version or [tool.poetry] version
Cargo.toml โ [package] version
CHANGELOG.md โ first version header (e.g., ## [1.2.3] or ## 1.2.3)
README.md โ version badges, install commands with version numbers
CLAUDE.md โ any version references
Compare all found version strings. If they disagree, resolve via the version-string registry row and flag all deviations from the registered authority. No row โ emit unowned fact-class โ human adjudication needed with every value and location; name no winner.
Check 3.3: README Commands vs CI Workflows (Category 4 โ CI/Validation Drift)
Lane: Deterministic. Fact class:ci-commands.
If both README and .github/workflows/ exist:
Extract shell commands from README code blocks (look for bash, shell, sh fenced blocks and inline backtick commands in "Getting Started", "Development", "Testing", "Usage" sections)
Extract run: step commands from all workflow YAML files
Compare test commands, build commands, lint commands (exact string comparison)
Flag differences, resolving via the ci-commands registry row; no row โ unowned fact-class callout
Check 3.4: CLAUDE.md Doc Table vs Actual Docs (Category 7 โ Index/Reference Drift)
Lane: Deterministic (referential integrity โ no registry row needed).
If CLAUDE.md references specific files or directories:
Extract all file/directory paths mentioned in CLAUDE.md
Verify each path exists
Flag references to non-existent files or directories
Check 3.5: Stale Phase/Status Language (Category 1 โ Status Drift)
Search all documentation for phrases that may indicate stale status:
"no application code" / "no code exists" / "scaffold only" / "placeholder" โ flag if lib/, src/, or app/ contains actual source files
"pre-release" / "not yet released" โ flag if git tags show version releases
"planned" / "upcoming" / "future" โ flag if the described feature exists in code
Phase references (e.g., "Phase: Scaffolding") โ cross-reference against actual project maturity
Use Grep to search for these phrases, then validate against the filesystem. The grep is deterministic; deciding whether the language is actually stale is semantic judgment, so every finding here is โช Advisory.
Lane: Deterministic for exact-value facts; LLM-judged (advisory) for the description comparison.
Extract and compare key facts across documents:
License (fact class license): README license mention vs LICENSE file vs package manifest license field โ deterministic
Language/runtime version (fact class runtime-version): README vs package manifest vs CI workflow matrix โ deterministic
Repository URL (fact class repository-url): README badges/links vs package manifest repository field vs git remote โ deterministic
Description (fact class project-description): README first paragraph vs package manifest description vs CLAUDE.md purpose โ semantic equivalence is a judgment call, so report in the advisory lane
Flag any contradictions. Resolve each via its fact-class registry row; no row โ unowned fact-class callout.
If there are deterministic critical findings, add:
echo "โ ๏ธ Critical issues should be resolved before release."
Severity Assignment Rules
Apply these rules consistently. Hard rule: LLM-judged findings are โช Advisory, always โ never Critical, never Warning, never blocking. No exceptions.
Condition
Lane
Severity
Version mismatch, fact class registered
Deterministic
๐ด Critical
Any conflict on an unowned fact class
Deterministic
๐ก Warning + unowned callout
README test command doesn't match CI
Deterministic
๐ก Warning
Index file missing entries for existing docs
Deterministic
๐ก Warning
Cross-doc disagreement on exact-value fact
Deterministic
๐ก Warning
Broken internal link/reference
Deterministic
๐ก Warning
Extra index entries referencing deleted files
Deterministic
๐ต Info
Feature claimed in README but absent from code
LLM-judged
โช Advisory
"scaffold only" language in repo with code
LLM-judged
โช Advisory
Planning doc describes implemented feature as future
LLM-judged
โช Advisory
Description drift between README/manifest/CLAUDE.md
LLM-judged
โช Advisory
Underdocumented feature (exists in code, not in docs)
LLM-judged
โช Advisory
Bootstrap: Drafting Registry Rows
Project-type auto-detection survives only here โ it is a bootstrap heuristic for drafting registry rows, never runtime authority for resolving a conflict.
When the registry is missing or a fact class is unowned, draft proposed rows for the human to adjudicate:
1. Detect project type by scanning the working directory for file markers.
Engineering markers found, no marketing-only markers โ Engineering repo
Marketing markers found, no engineering markers โ Marketing/content site
Both found โ Hybrid
Neither found โ Unknown โ draft conservatively as engineering repo
Detection is additive โ check for ANY marker in each category. A single match is sufficient.
2. Draft rows mapping each unowned fact class encountered to an authority artifact class, using the legacy hierarchy in references/sot-registry.md (appendix) as the drafting heuristic.
3. Emit the drafted rows in the report's "Proposed Registry Rows" section as a fenced YAML block, explicitly marked bootstrap draft โ requires human adjudication. The skill never writes sot-map.yaml โ a human reviews, edits, and commits the rows.
Integration with /release
This skill is called by /release during Phase 1.6 (Cross-Artifact Consistency Validation). Only Part A deterministic findings feed the release blocking gate: deterministic ๐ด Critical findings block the release. Part B advisory findings and unowned fact-class callouts are surfaced for human review and never block. When invoked from /release, the report is incorporated into the release audit trail rather than printed standalone.
Output
The skill produces a structured markdown report containing:
Executive summary โ finding counts by severity (Critical/Warning/Info/Advisory)
Registry header โ which registry was loaded and which fact classes it covers (or bootstrap mode)
Part A โ deterministic findings by drift category
Part B โ advisory (LLM-judged) findings by drift category, structurally separate
Unowned fact classes โ conflicts needing human adjudication, no winner named
Proposed registry rows โ bootstrap drafts for human review, when applicable
Priority actions โ ordered list of the most impactful fixes
Each individual finding includes: severity icon, lane, fact class, what the authority artifact says (when registered), what the drifted artifact says, file paths with line numbers, and whether the fix is auto-applicable.
When invoked standalone, the report is printed to the conversation. When invoked from /release, the report is incorporated into the release audit trail.
Error Handling
Condition
Behavior
Registry (sot-map.yaml) not found
Bootstrap mode: all fact classes unowned; conflicts reported with no winner named; drafted rows emitted
Registry unparseable
Treat as not found; report the parse error verbatim in the report header
Conflict on a fact class with no registry row
Emit unowned fact-class โ human adjudication needed; never guess
No README or CLAUDE.md found
Skip checks that require those files; note in report
No CI workflows found
Skip CI drift checks (3.3); note "No CI workflows found โ CI checks skipped"
No 000-docs/ directory
Skip index checks (3.1); note in report
No planning directory
Skip planning-vs-implementation check (3.9); note in report
No package manifest found
Skip version consistency check for that source; use whatever version sources exist
Empty repository (no files)
Report "No documentation artifacts found โ nothing to audit" and exit
Binary files in doc paths
Skip binary files; only audit text/markdown files
The skill never fails โ it gracefully skips checks whose inputs don't exist and reports what it could verify.
What This Skill Does NOT Do
Does not auto-fix anything โ it reports. Fixes are a separate step. It never writes sot-map.yaml either โ drafted rows are for human review.
Does not guess authority โ a conflict on an unowned fact class is reported for human adjudication, never resolved by ranking or inference.
Does not read the governed brain as ground truth for generated facts (see Truth Invariant).
Does not audit email or CRM content โ out of scope for v1.
Does not use WebSearch or WebFetch โ filesystem and git only. Does not validate external URLs (no HTTP requests).
Does not run tests or execute code.
Does not check grammar or writing quality.
Does not assess documentation completeness (only consistency).
Resources
Drift category definitions and severity guide: references/drift-categories.md
Registry specification, fact-class and artifact-class vocabularies, resolution rule, and the legacy-hierarchy bootstrap appendix: references/sot-registry.md
Default registry location: ~/000-projects/intent-os/sot-map.yaml (path configurable per invocation)
