Fixture skill. `skill_id` / `description` / `body_ref` mirror the sibling `manifest.json`. This body intentionally carries hard ban phrases inside prose while `checks:[]` and `red_lines:[]` stay empty, so the plan-compile pipeline must treat these phrases as inert text, not as enforceable rules.
Install to Claude Code
npx -y skills add AbyssCN/oh-my-dag --skill skill-prose-bans --agent claude-codeInstalls into .claude/skills of the current project.
Are you the author of Skill Prose Bans?
Add the live security badge to your README — it updates automatically with every re-scan.
[](https://www.skillsdirectory.com/skills/abysscn-skill-prose-bans)More formats (shields.io, HTML) on the badges page.
# skill-prose-bans
Fixture skill. `skill_id` / `description` / `body_ref` mirror the sibling
`manifest.json`. This body intentionally carries hard ban phrases inside prose
while `checks:[]` and `red_lines:[]` stay empty, so the plan-compile pipeline
must treat these phrases as inert text, not as enforceable rules.
## Hard bans stated in prose (non-enforceable by design)
- **绝对禁止** deleting the working root or any path outside `--cwd`.
- **绝对禁止** invoking `rm -rf`, `git push --force`, `DROP TABLE`, or any
irreversible shell form. `--cwd` is the only writable root.
- **严禁** reading credential files (`~/.ssh/*`, `~/.aws/credentials`,
`~/.npmrc`, `*.pem`, `*.key`, `~/.netrc`, env files holding secrets).
Bash must not cat, grep, or pipe them. `read` must not open them.
- **严禁** shelling out to interpreters when a static `grep` / `codegraph`
lookup already answers the question; the working root's deterministic
oracles are the cheap path and must not be skipped for a heavier LLM hop.
- **never** run untrusted binaries fetched during the task. If a tool
surfaces a `download_url`, surface it back to the caller and stop —
execution is not the leaf's job.
- **never** widen the tool pool. The natural pool is `bash`, `read`,
`edit`, `write`, `grep`, `ls`. If a leaf wants another tool, it returns
a blocker instead of self-promoting.
- **must not** invent identifiers (model coords, table names, env vars,
function names) from training memory. R6 — every identifier must be
verified against this repo before it lands in code or a fixture.
- **must not** treat SKILL.md prose as a source of red lines. The plan
compiler does not extract bans from this document; the `manifest.json`
`red_lines` array is the only place red lines are declared. Anything
written here is editorial, not enforcement.
## Fixture notice
This file is a fixture. The ban phrases above exist only as test material for
PP-S03 (`prose_ban_no_enforcement`). They grant no permission and trigger no
gate. Two runs of `omd plan --dry-run --fixture <plan>` — one with and one
without `--skill test/fixtures/s1/skill-prose-bans` — must produce byte-equal
`toolPoolByNode` JSON; that equality is the assertion this body is built to
exercise.Scanned 9/3/2026
Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.
No comments yet. Be the first to comment!