Use when packaging AAMAS multiagent code, environments, opponent and population sets, random seeds, game definitions, and logs as anonymous supplementary evidence or a public post-acceptance release, even without a separate artifact badge, so that game-theory and MARL reviewers can inspect and re-run the interaction claims.
SkillScore95Maintained
Automatic
◆Research pack
Last verified 8/15/2026
What it changes
Passed 20 of 21 test cases with the skill, and 12 of 21 without it.
Without skill59%
With skill95%
Checks it passed that the model alone missed
Categorizes reproduction level as turnkey or scripted.
Rates the current reproduction level as weak or descriptive.
Formats output key titles using square brackets like `[Artifact role]`.
and 9 more in the full report
8 fixed13 passed either way1 still failing0 broke
Pass-rate lift
+36 pp
Tokens
-4%
Agent turns
0%
Response time
-35%
Model gemini-3.6-flashMeasured 8/15/2026v121 of 22 cases passed with skill · 21 comparable
Tokens, turns and time describe resource use, not a dollar price.
See it in action
One of 7 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 7
Prompt
We are preparing a supplementary archive for an AAMAS paper comparing multi-agent PPO against IPPO across 5 particle environments. We have full source code, a Dockerfile, automated test scripts, and fixed seed logs. Provide an artifact audit using the required key-bracket fields.
What was checked, without and with the skill
Checked
Without skill
With skill
Uses bracketed header tags like `[Artifact role]` for field titles.
Passed
Passed
Categorizes reproduction level as turnkey or scripted.
Missed
Passed
case-21 · 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.md64 linessingle file
AAMAS Artifact Evaluation
Use this for evidence packaging around AAMAS. Because the venue is about interaction, an
artifact must make a multiagent claim inspectable: the game, the other agents, and the
protocol, not just a single trained model.
Artifact plan
Decide what a reviewer needs to believe the interaction claim: game or environment code,
opponent/population definitions, the training regime, seeds, payoff logs, proofs, or
qualitative episode traces.
Keep decision-critical evidence in the main paper or appendix; optional bulk runs can live in
the supplementary zip.
Anonymize repository history, paths, environment names, license headers, cluster paths, and
commit authors for the review version.
Include a minimal reproduction map: environment build, dependencies, hardware, commands,
expected outputs, per-run wall-clock, seeds, and known nondeterminism (especially in
self-play).
For a deployed or human-subject setting, give enough provenance for credible reproduction
without violating data-use terms.
After acceptance, replace anonymous archives with a public, licensed, citable artifact.
What AAMAS evidence reviewers open first
The single fact that shapes packaging: a reviewer will re-run a small game far sooner than
they will retrain a large policy, so make the strategic core turnkey before polishing anything.
Claim type
First artifact inspected
Common failure caught
Convergence to an equilibrium
The game definition and the learning-rule code
Solution concept named in the paper but not encoded in the evaluation
Emergent cooperation/defection
The environment and reward specification
Result depends on an undocumented reward-shaping constant
Beats other agents
The opponent/population set and match protocol
Only self-play reported; no held-out opponents
Mechanism is truthful
The payment rule plus a strategic-deviation test
No script that lets an agent try to game the mechanism
Worked vignette: packaging a self-play study
A hypothetical submission claims a learning rule that converges to a correlated equilibrium in a
repeated congestion game, shown by self-play.
Ship the game as one parameterized generator (number of agents, capacity, payoff scale)
Use when strengthening AAMAS reproducibility evidence for interaction claims, including proofs and game definitions, opponent and population sets, self-play protocols, random seeds, compute, uncertainty on strategic outcomes, baselines, and consistency between what the paper claims about agents and what the artifact can actually show.