docs-update

Update Documentation

Synchronize documentation with the current codebase state: one file, a category, or everything in docs/.

Anti-Hallucination Guidelines

Every statement written must trace back to a file read or grepped this run:

1. Verify each claim before writing it, read the actual source, don't recall it
2. Get counts from find/grep, never estimate ("5 services" means 5 files were counted this run)
3. Delete claims about features that no longer exist rather than leaving them stale
4. After writing, re-read the diff and drop anything not backed by evidence gathered this run

Workflow

Phase 1: Resolve Scope

Parse the invocation argument:

• No argument or all: update every doc that exists in docs/
• <doc-name> (e.g. architecture): resolve via the mapping below, update only that file
• category:<name> (core, data, infrastructure, development): update every doc in that category

Argument	File Path	Category
architecture	docs/architecture.md	core
onboarding	docs/onboarding.md	core
data-model	docs/data-model.md	data
deployment	docs/deployment.md	infrastructure
security	docs/security.md	infrastructure
contributing	docs/contributing.md	development

Existence check: stop before doing anything else if the target is missing. This skill only updates docs that already exist; it never creates one.

• all or category:<name>: silently exclude any mapped path that isn't on disk (e.g. find docs/*.md and intersect with the category's paths); the run proceeds with whatever remains
• <doc-name>: resolve the path, then check it with test -f <resolved-doc-path> (or Read). If it does not exist, STOP the entire skill run right here: do not read further, do not run Phase 2-5, do not Write or Edit that path under any pretext (not as a "special case," not with a caveat about verified content). Tell the user the doc doesn't exist yet and point them to docs-init, which owns creation and the canonical templates. This is the final action for that run.