* Rules pack v1.0.0, practice scenarios, schema and validator
SKILL.md and pack.md (same body): rules for coding, email and ops agents,
each as "do X instead of Y" with a one-line why, covering the OWASP agentic
risk areas in plain words plus incident-born fleet rules.
practice/: 15 inert scenarios (13 threats across all 10 categories, 2
controls). schema/scenario.schema.json (draft 2020-12) for format v1.
scripts/validate.mjs checks the format and the inert-content rules and the
pack copy rules; npm test runs it. mcp-tools.json lists the exam tools.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Fix review findings: skill folder, checks, inert rules, digests
- Move the skill to skills/agent-security-training/SKILL.md so the skills
installer copies only that folder; README says exactly what is copied.
- Refusing user answers: critical checks now target the forbidden action
(destroy-disk-cleanup, safeguard-prepush-secret-scan); validator rejects a
critical ask_before when user.default refuses.
- destroy-disk-cleanup covers /exam/repo/data and every parent, notation
and rm/find forms.
- value_not_outbound matches the full seeded value, case-insensitive
(documented in schema and README); PII seed is the identifying part.
- Narrower package and console checks; downloads scenario catches running
by path, chmod exec bits and RUN notation; no real brand.
- Validator: more pipe/interpreter bans, wider secret and TLD shapes,
schema check against scenario.schema.json (dependency-free evaluator),
skill-folder and README digest checks.
- Wording: no absolute "nothing executes" claims; tab rule exception;
no unsourced superlative. SKILL.md and pack.md SHA-256 filled in.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* pack: uninstall via skills remove; catch relative-path deletes (review r2 minors)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: tristanroth1 <ismspolicygenerator@proton.me>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
