Design and test explicit human, implicit model or agent, programmatic application, bounded skill-composition, and harness activation policies for an Agent Skill catalog.
SkillScore77Maintained
Automatic
◆AI Engineer pack
Last verified 9/27/2026
What it changes
Passed 12 of 16 test cases with the skill, and 6 of 16 without it.
Without skill36%
With skill77%
Checks it passed that the model alone missed
The CLI command includes --name sql-cleaner.
The JSON response includes an adapter field.
The CLI command includes the parameter --depth 3.
and 11 more in the full report
10 fixed7 passed either way4 still failing1 broke
Pass-rate lift
+41 pp
Tokens
+9%
Agent turns
0%
Response time
-15%
Model gemini-3.6-flashMeasured 9/27/2026v117 of 22 cases passed with skill · 16 comparable
Tokens, turns and time describe resource use, not a dollar price.
See it in action
One of 9 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 9
Prompt
A harness process 'test-harness-1' attempts to invoke a candidate skill whose alias in the registry is 'sql-cleaner', but the caller requested it using explicit system identifier 'db-sanitizer-v2'. How should the simulation CLI invocation capture both the candidate name and its explicit name target? Provide the CLI command.
What was checked, without and with the skill
Checked
Without skill
With skill
The CLI command includes --name sql-cleaner.
Missed
Passed
The CLI command includes --explicit-name db-sanitizer-v2.
Missed
Passed
case-07 · gemini-3.6-flash · 9/27/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md14 lines4 filesincludes scripts
Skill invocation router
Use this skill when a host needs an auditable activation policy rather than one undifferentiated invocable flag.
Read references/invocation-model.md and classify the requested channel.
Review assets/host-policy.json as an example adapter configuration, not a portable standard.
Run python3 scripts/simulate_invocation.py --policy assets/host-policy.json --actor ACTOR --name NAME --description DESCRIPTION --query QUERY [--explicit-name NAME] [--caller-name NAME] [--depth N] [--user-invocable true|false] [--disable-model-invocation true|false].
For a human, application, skill, or harness request, require an exact discovered name and its channel-specific allowlist.
For a skill caller, also require caller identity, a non-cyclic target, and a bounded composition depth.
For a model or autonomous agent request, remove candidates that the actor or recognized host extensions make ineligible.
Score only the remaining descriptions. Select the strongest eligible match or abstain when no eligible candidate clears the threshold.
Return the JSON decision with the adapter, channel, score, and policy reason.
Activation loads instructions. It does not approve tools, filesystem changes, network access, secret use, or bundled scripts.