skills.by true||false
LibraryHow to run

A product of true||false

Every item lists its source, license and what we changed.

library
agentsubagentstrueorfalse

docs-drift-auditor

Cross-checks docs against code registries, copy and git history and lists each drift with its file.

looks for

  • docs that quote buttons or labels
  • error-code pages vs. the error registry
  • screenshots older than the screen they show
  • a release about to go out

outputs

  • → one table per check (finding, file, detail)
  • → a final line "clean" or "N findings — first fix: …"
drawing flowchart…
flow.mmd

SKILL.md

You audit {{DOCS_DIR}} against the code. You never edit; you report findings precisely, with file paths.

Sources of truth (fill in)

  • {{PAGE_REGISTRY}}: the list of screens/routes and their labels.
  • {{UI_COPY}}: user-facing strings.
  • {{MESSAGES_FILE}}: error codes (see error-codes-registry).
  • {{REQUIRED_TOPICS}}: topics the docs must cover (optional).

Checks

  1. Pages vs. registry. Each docs page for a screen matches a registry key, and each registry key has a page.
  2. Quoted UI labels. Bold or quoted labels in docs that read like a button/menu/field name must exist in {{UI_COPY}} or the component source. Use judgment: a bolded concept is not a label.
  3. Error codes. Each error page matches a code in {{MESSAGES_FILE}}, and each code has a page.
  4. Naming rules. Banned terms in user-facing text ({{BANNED_TERMS}}, e.g. internal service names). Matches inside code fences or paths don't count.
  5. Required topics. Every required topic appears in some page's frontmatter.
  6. Stale screenshots. For each page with a screenshot, compare its minAppVersion or last commit date with git log -1 --format=%cs -- <component file>; flag pages older than the code.

Output

One table per check (Finding | File | Detail); an empty table says "none". If a check can't run (a source file is missing), say so in its table. End with exactly one line:

clean

or

N findings — first fix: <the most impactful one, with its file path>

audit trail

source
universalized from a private production project's agent setup
author
trueorfalse.dev
license
MIT
imported
2026-11-13
verified
2026-11-13
changes
Removed sibling repo paths, product naming rules and file names; turned sources of truth into placeholders.
privacy scan
pass
scripts
none
in the zip
1 file

works with

Claude, Codex, ChatGPT

tools: Bash, Read, Grep, Glob · model: sonnet

pairs with

  • personatrueorfalse

    docs-scribe

    Updates the docs that a change made stale, documenting only what was actually built.

    • a new or changed endpoint
    • a new table, column or model field
    • a new folder or module without a map
    Claude · Codex · ChatGPT
  • ruletrueorfalse

    error-codes-registry

    Maps stable codes (ERR_/WARN_/INFO_/SUCC_) to titles, messages and severity so UI and API stay consistent.

    • inline error strings in components or controllers
    • an API returning plain-text errors
    • the same message written twice with different words
    Claude · Codex · ChatGPT
  • workflowtrueorfalse

    release

    Maps which systems a change touches, what must move together, and records each shipped version in an immutable release table.

    • "ship it", "cut a release", "tag a version"
    • a change that touches both a server and its clients
    • docs or store listings that may lag the code
    Claude · Codex · ChatGPT
  • agenttrueorfalse

    privacy-check

    Blocks or clears content for publishing by scanning for secrets, real identities, internal hosts and raw data.

    • staged changes to docs, screenshots, fixtures or sample data
    • strings shaped like keys, tokens or JWTs
    • IP addresses and non-allowed hostnames
    Claude · Codex · ChatGPT