strata-lint

name	strata:lint
disable-model-invocation	true
description	Scan vault or staged files for credentials, PII, and (opt-in) PHI via pluggable JSON presets. Use when the user says "check for secrets", "scan for credentials", "is this safe to save", or mentions an API key/token/password/patient data. Runs automatically as a pre-step inside /strata:save, /strata:decide, and /strata:correct, so secret-scanning never depends on a probabilistic skill pick.

strata:lint

Pluggable scanner. Patterns live in presets/*.json, credentials by Default, PII and region-specific PHI as opt-in presets. Add your own.

Even though the vault lives outside the repo and isn't usually committed, it is sync'd to other team members, same blast radius. Lint before sync.

When to use

• Periodically across the whole vault (--scope vault, the default).
• Before committing if you're export-promoting a decision into the host repo (--scope staged).
• As part of a pre-push git hook in the host repo (recommended — see INSTALL.md).

How

The front door handles the venv automatically, invoke via ${CLAUDE_PLUGIN_ROOT}/bin/strata:

# Default: scan the whole vault with the `secrets` preset
"${CLAUDE_PLUGIN_ROOT}/bin/strata" lint

# Add region-specific PHI presets
"${CLAUDE_PLUGIN_ROOT}/bin/strata" lint \
  --preset secrets,pii,phi-us

# Staged *.md files in the host repo
"${CLAUDE_PLUGIN_ROOT}/bin/strata" lint \
  --scope staged --preset secrets,pii

# A specific file or directory
"${CLAUDE_PLUGIN_ROOT}/bin/strata" lint \
  --scope ~/StrataVault/myrepo/pr-context/feat-x/

# Treat warnings as errors (CI mode)
"${CLAUDE_PLUGIN_ROOT}/bin/strata" lint \
  --scope vault --preset secrets,pii --strict

Exit code 0 = clean, 1 = findings, 2 = misuse.

Bundled presets

Preset	Catches
secrets (default)	AWS, GitHub, Slack, OpenAI, Anthropic tokens, JWTs, PEM blocks, connection-string passwords, Azure account keys, GCP service-account markers
pii	Credit cards (Luhn-validated), client/user-id literal assignments, emails
phi-uk	NHS numbers (Mod-11 verified), UK postcodes (BS7666)
phi-us	SSN (with structural validation), DEA numbers (checksum-verified)

All findings are BLOCK-level by default, they detect actual identifier values, not vocabulary. The lint deliberately does not warn on words Like "SSN", "NHS number", or "HIPAA" appearing in prose, because team memory Notes will legitimately discuss systems that handle that data ("the SSN flow Goes through service X", "we don't store NHS numbers in this table"). That's the whole point of ADRs.

Use --strict in CI to fail on any warn-level finding from presets that have Them (currently only pii does).

Tuning / adding presets

Each preset is a JSON file in presets/:

{
  "id": "my-org",
  "description": "Org-specific patterns",
  "block": [
    { "name": "internal-token", "regex": "MYORG-[A-Z0-9]{32}" }
  ],
  "warn": [
    { "name": "internal-host", "regex": "\\binternal\\.myorg\\.example\\b" }
  ]
}

Optional flags (i, m, s, case-insensitive, multi-line, dotall) and validator (a function name from VALIDATORS in memory-lint.py — currently nhs_mod11, luhn, us_ssn, dea_checksum).

Reference it via --preset secrets,my-org. PRs adding new region presets Welcome.