security-boundary-review
Skill22 starsChanged 11 months agoApache-2.0View on GitHub
Review a change touching a security/tenancy boundary, SSRF/CSP/auth guard, or a doc claiming a security control. Use when: reviewing or testing a guard, writing security docs, planning a boundary slice.
2 other projects use a version of this.
Adopted by
Most recently taken up by
- Xakki/ai-agents-skillstheir own version, first seen Nov 8, 2026
- timfewi/unix-agent-skillstheir own version, first seen Oct 17, 2026
Other versions2 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 2
- Xakki/ai-agents-skills★ 6same name
- timfewi/unix-agent-skills★ 2same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository7 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- daemon-debugSkill
- daemon-health-checkSkill
- flake-triageSkill
- messaging-queueSkill
- searchSkill
The file
We split this skill into the rules it gives an agent
10 rules. About 0 are common, 8 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈8
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 12
Read the cited mechanism: Do not trust the description or the spec.
- Distinctive: line 14
Assert the refusal's provenance: Ask: "If I deleted this control, would this assertion still pass?" In a layered path it usually would.
- Very specific: line 18
Assert coincidence premises: If a test depends on equal digests, colliding keys or the same timestamp, assert that premise first with …
- Distinctive: line 26
When CSP-adjacent behaviour is enforced at two layers (validator or renderer first, then the header), state which one refuses first.
- Distinctive: line 34
Reviewing: verify the mechanism before you write the hint.
- Distinctive: line 38
Headline coverage: check that the slice set covers the DoD's headline capability.
- Very specific: line 40
Integration points exist: verify that named integration points exist and sit on the claimed path.
- Distinctive: line 41
Impossible slice: if a slice cannot be done as written, STOP and return a re-scope report.
- Distinctive: line 42
Traceability table as tie-breaker: when the spec and a decision appear to conflict, consult the spec's traceability table before escalating.
- Distinctive: line 44
Hazards for future consumers: when you spot a hazard for a consumer that does not exist yet, hand it forward as an explicit constraint on …
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Security boundary review
A green suite and the author's self-review are necessary, not sufficient. Both rewinds on the publish epics landed on code that was already green. Attack the boundary the author's tests assume. Code-level rules live in .claude/rules/publish-security-review-lessons.md.
Reviewing a guard or boundary change
- Read the cited mechanism. Do not trust the description or the spec. Example: a claim of "replaces upstream CSP" pointed at a strip-merge that kept hostile directives.
- Attack the collision path, not the naming path. Tests often prove that you cannot name another tenant's resource. Also test what happens when two tenants' keys, digests or hashes collide.
- Assert the refusal's provenance. Ask: "If I deleted this control, would this assertion still pass?" In a layered path it usually would. Assert both of these:
- the error names the control's verdict, e.g.
"upstream refused"; - the dangerous side effect was never reached, e.g. the dialer was called exactly once.
- the error names the control's verdict, e.g.
- Mutation-verify once. Revert the fix or break the guard, rerun, and confirm the test fails on a behavioural assertion rather than on a key-comparison line. Restore the tree afterwards.
- Assert coincidence premises. If a test depends on equal digests, colliding keys or the same timestamp, assert that premise first with
t.Fatalf("premise broken: ..."). - Never pin a value by the symbol that produces it. Spell the expected security constant as a literal. A test that compares against
proxiedUpstreamStyleSrcpasses any mutation of it. - Name the field you mean. Assert on the directive or field itself, e.g.
script-srcsources are all hashes. A document-wide substring such as "nounsafe-inlineanywhere" breaks the day an unrelated directive legitimately gains it.
Security docs
- A sentence describing a shipped control's scope is load-bearing. Grep the guard's production callers and check the claim against the call sites. The count is usually one.
- The tell is coverage words: "every", "all", "the same guard also".
- When CSP-adjacent behaviour is enforced at two layers (validator or renderer first, then the header), state which one refuses first.
- One false claim obliges a whole-doc sweep for that claim's shape. Fixing only the cited line blesses its siblings.
- Fix a false claim by naming the real gap, the silent failure it hides, and the tracking task id. Never delete the sentence.
Briefs and fix_hints
- Dispatching: mark your understanding as a hypothesis, and tell the subagent to verify it against the code and report contradictions.
- Reworking: a fix_hint's mechanism claim is a diagnosis, not a spec. Follow the code, and record any divergence in the commit body.
- Reviewing: verify the mechanism before you write the hint.
Planning a boundary slice
- Headline coverage: check that the slice set covers the DoD's headline capability. If the decomposition narrows it, write the narrowing down as a scope decision.
- Owning file in scope: a criterion that names a test tier, config, doc or generated asset needs its owning file in the task's
fileScope. For example, real-Chrome assertions need the*_e2e_test.gofile with thechromee2etag. - Integration points exist: verify that named integration points exist and sit on the claimed path. For the public plane, check
moduleRoleandrolePublicModulesmembership. - Impossible slice: if a slice cannot be done as written, STOP and return a re-scope report. Never widen scope silently or fake a substitute.
- Traceability table as tie-breaker: when the spec and a decision appear to conflict, consult the spec's traceability table before escalating.
- Irreversible batch operations: phrase the criterion structurally, e.g. "the destructive call is unreachable from a partial scan, verified by reading every call site."
- Hazards for future consumers: when you spot a hazard for a consumer that does not exist yet, hand it forward as an explicit constraint on that consumer's task.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

