Use before routing a /meta-agent request to choose single-agent-creator, team-builder, or agentlas-packager from the user's wording and available files.
SkillScore100Maintained
Automatic
◆AI Engineer pack
Last verified 8/15/2026
What it changes
Passed 22 of 22 test cases with the skill, and 13 of 22 without it.
Without skill59%
With skill100%
Checks it passed that the model alone missed
Selects team-builder mode
Sets ontology_backed to true
Selects agentlas-packager mode
and 6 more in the full report
9 fixed13 passed either way0 still failing0 broke
Pass-rate lift
+41 pp
Tokens
+5%
Agent turns
0%
Response time
-47%
Model gemini-3.6-flashMeasured 8/15/2026v122 of 22 cases passed with skill
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
We have a folder of YAML agent configurations from a previous framework that we need to import and clean up for our current system. What mode applies?
What was checked, without and with the skill
Checked
Without skill
With skill
Selects agentlas-packager mode
Missed
Passed
case-16 · gemini-3.6-flash · 8/15/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md54 linessingle file
Mode Classification
Pick one Agentlas meta-agent mode before generating or repairing files.
Procedure
Inspect the user request and any provided path, repo, ZIP, prompt, or agent
files.
Step 0 - existing material wins: if existing material is being converted,
repaired, cleaned, imported, or released, choose agentlas-packager.
Step 1 - count independent ownership boundaries. Ask how many roles must
independently own all three of:
their own memory/context;
their own tools/permissions;
their own success criteria.
One boundary means single-agent-creator. Two or more boundaries means a
team-builder candidate. If the boundary count is unclear, run the clarify
question loop before generating; do not infer from the word "team" alone.
Step 2 - check synthesis need for multi-boundary candidates. If those role
outputs must be routed, reviewed, synthesized, or chained through
produces/consumes dependencies, choose team-builder and require an
orchestrator/HQ plus memory, policy, eval, and QA. If the roles are
unrelated, create separate single-agent packages instead of one team.
Step 3 - shape guard. single-agent-creator may have many skills/tools but
must not emit multiple loose worker agent.md files. team-builder may be
small, but it must not omit the orchestrator/HQ.
Use keyword signals only as hints after the ownership-boundary check:
MULTI hints: separate memory partitions, tools or permissions that must
not be merged, role-to-role review/policy separation, and
produces/consumes pipelines.
SINGLE hints: one coherent job, many tools/skills owned by one worker, no
routing or final synthesis requirement.
Overlay check: if the request depends on knowledge search over user
documents, evidence-based or citation-attached generation, or a document
corpus (HWPX/docx/pdf/제안서/계약서/견적서), additionally apply the
ontology-backed-agent overlay (modes/ontology-backed-agent.md) with
ontology_backed: true on the chosen base mode.
Loop policy: derive loop_policy from task purpose and risk using
.agentlas/contract-injection-map.json risk tiers — none for simple
one-shot tasks, self-correct for complex or long-running work, verified