Skip to content

docs(claude_skill): skill-safety rationale (CSKILL-001..011)#15

Merged
jhumel-code merged 1 commit into
mainfrom
feat/claude-skill-rationale
Jun 8, 2026
Merged

docs(claude_skill): skill-safety rationale (CSKILL-001..011)#15
jhumel-code merged 1 commit into
mainfrom
feat/claude-skill-rationale

Conversation

@jhumel-code

Copy link
Copy Markdown
Collaborator

Adds the rationale doc for the claude_skill rule pack. (The engine resolves rules from trustabl-rules; the rulebook is the defensible grounding — threat model, severity/confidence defense, OWASP LLM Top-10 references.)

Covers CSKILL-001/002/003/010/011/020/040/050:

  • bundled-script egress / credential-read (010/011) — the payload-in-aux-file surface body-only scanners miss
  • hidden-Unicode coverage on 040 (Unicode Tags block U+E0000–E007F + bidi overrides)

POLICY_INDEX.md regenerated (adds the 8 CSKILL rows + table reflow; no other content change).

Pairs with the engine PR and the trustabl-rules PR (CSKILL packs). check_rulebook.py validates the doc against the shipped rules, so merge together with / after the rules PR.

Refs: TR-218, TR-219

Rule-by-rule defense + OWASP LLM Top-10 mapping for the claude_skill pack:
CSKILL-001/002/003 (auto-approved shell, pre-model dynamic-context exec +
egress/secret exfil), 010/011 (bundled-script egress + credential read), 020
(external content), 040 (injection markers incl. hidden-Unicode), 050
(model-invocable + side-effecting). Regenerate POLICY_INDEX.md.

Refs: TR-218, TR-219
@jhumel-code jhumel-code merged commit 1937f61 into main Jun 8, 2026
3 of 4 checks passed
@jhumel-code jhumel-code deleted the feat/claude-skill-rationale branch June 8, 2026 07:52
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant