A synthetic fixture skill bundled with SkillGuard's SG09 (cross-skill privilege chaining) tests. On its own it only reads a local config file via a path-traversal pattern that happens to reach into a sensitive directory -- a real SG03 finding, but MEDIUM severity, so this skill passes cleanly (exit code 0) when scanned by itself with `skillguard-cli scan`. It is inert: no real filesystem or network calls. Do not install or run this in a real agent.
SkillScore100Maintained
Automatic
◆Security & AppSec pack
Last verified 8/3/2026
What it changes
Passed 23 of 23 test cases with the skill, and 17 of 23 without it.
Without skill74%
With skill100%
Checks it passed that the model alone missed
Identifies the rule code as SG09
Recommends declaring per-skill sandboxing
Identifies vault-reader as one of the skills
and 5 more in the full report
6 fixed17 passed either way0 still failing0 broke
Pass-rate lift
+26 pp
Tokens
-22%
Agent turns
0%
Response time
-30%
Model gemini-3.6-flashMeasured 8/3/2026v123 of 23 cases passed with skill
Tokens, turns and time describe resource use, not a dollar price.
See it in action
One of 6 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 6
Prompt
When evaluating LLM agent skills, static security tools flag scenarios where one tool accesses sensitive local files and another has network egress. Some developers assume this is rule SG01 or SG05. What is the specific SkillGuard finding code for cross-skill privilege chaining?
What was checked, without and with the skill
Checked
Without skill
With skill
Identifies the rule code as SG09
Missed
Passed
case-03 · gemini-3.6-flash · 8/3/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md16 linessingle file
Vault Reader (SG09 fixture)
Paired with the sibling report-uploader skill in this same directory:
neither skill trips a HIGH finding on its own, but scanning the two
together with skillguard-cli scan-set should flag SG09 -- cross-skill
privilege chaining -- because this skill can reach a sensitive file and
report-uploader can send data over the network, and neither declares
per-skill sandboxing.
Cross-validate CLI flags, docs, tests, and targets for consistency across the codebase. Use this skill whenever the user asks to: audit the codebase, check for consistency issues, find undocumented flags, verify test coverage, validate targets.yaml, check handler split conventions, or verify oplog instrumentation. This is a read-only audit — it reports issues but never modifies files. Use after large refactors, before releases, or whenever you suspect docs/code/tests have drifted out of sync.