MCP HubMCP Hub
Retour aux compétences

test-team-coordination

pjt222
Mis à jour Yesterday
3 vues
17
2
17
Voir sur GitHub
Métaaitestingdesign

À propos

Cette compétence exécute des scénarios de test prédéfinis sur des équipes d'IA pour valider et comparer leurs modèles de coordination. Elle observe les comportements des équipes, évalue les critères d'acceptation et génère un fichier structuré RESULT.md contenant les scores et l'analyse. Utilisez-la pour valider la coordination attendue, comparer les modèles sur des charges de travail équivalentes ou établir des performances de référence pour les compositions d'équipes.

Installation rapide

Claude Code

Recommandé
Principal
npx skills add pjt222/agent-almanac -a claude-code
Commande PluginAlternatif
/plugin add https://github.com/pjt222/agent-almanac
Git CloneAlternatif
git clone https://github.com/pjt222/agent-almanac.git ~/.claude/skills/test-team-coordination

Copiez et collez cette commande dans Claude Code pour installer cette compétence

Documentation

Test Team Coordination

Execute a test scenario from tests/scenarios/teams/ against the target team. Observe coordination pattern behaviors, evaluate acceptance criteria, score the rubric, and produce a RESULT.md in tests/results/.

When to Use

  • Validating that a team's coordination pattern produces expected behaviors
  • Running a structured test after modifying a team definition or agent
  • Comparing coordination patterns by running the same scenario with different teams
  • Establishing baseline performance metrics for a team composition
  • Regression testing after adding new agents or changing team membership

Inputs

  • Required: Path to the test scenario file (e.g., tests/scenarios/teams/test-opaque-team-cartographers-audit.md)
  • Optional: Run ID override (default: YYYY-MM-DD-<target>-NNN auto-generated)
  • Optional: Team size override (default: from scenario frontmatter)
  • Optional: Skip scope change (default: false — inject scope change if defined)

Procedure

Step 1: Load and Validate Test Scenario

1.1. Read the test scenario file specified in the input.

1.2. Parse YAML frontmatter and extract:

  • target — the team to test
  • coordination-pattern — the expected pattern
  • team-size — number of members to spawn
  • Acceptance criteria table
  • Scoring rubric (if present)
  • Ground truth data (if present)

1.3. Verify the scenario file has all required sections:

  • Objective
  • Pre-conditions
  • Task (with Primary Task subsection)
  • Expected Behaviors
  • Acceptance Criteria
  • Observation Protocol

Got: Scenario file loads, parses, and contains all required sections.

If fail: If the file is missing or unparseable, abort with an error message identifying the missing file or malformed section. If optional sections (Rubric, Ground Truth, Variants) are absent, note their absence and continue.

Step 2: Verify Pre-conditions

2.1. Walk through each pre-condition checkbox in the scenario.

2.2. For file-existence checks, use Glob to verify.

2.3. For registry count checks, parse the relevant _registry.yml and compare total_* against actual file counts on disk.

2.4. For branch/git state checks, run git status --porcelain and git branch --show-current.

Got: All pre-conditions are satisfied.

If fail: If any pre-condition fails, record it as BLOCKED in the results. Decide whether to proceed (soft pre-condition) or abort (hard pre-condition like missing target team file). Document the decision.

Step 3: Load Coordination Pattern Criteria

3.1. Read tests/_registry.yml and locate the coordination_patterns entry matching the scenario's coordination-pattern value.

3.2. Extract the key_behaviors list for this pattern.

3.3. These behaviors become the observation checklist — each must be watched for during execution and recorded as observed/not observed.

Got: Pattern key behaviors loaded and ready for observation.

If fail: If the coordination pattern is not defined in the registry, use the scenario's Expected Behaviors section as the sole observation source. Log a warning.

Step 4: Execute the Task

4.1. Create the result directory: tests/results/YYYY-MM-DD-<target>-NNN/.

4.2. Record T0 (task start timestamp).

4.3. Read the target team's definition from teams/<target>.md, extract the CONFIG block, and activate the team: call TeamCreate with the team name, spawn teammates using each member's subagent_type, and create tasks from the CONFIG tasks list. Use the team-size from the scenario. Pass the Primary Task prompt verbatim from the scenario's Task section.

4.4. Observe the team's execution phases. Record timestamps for:

  • T1: Form assessment / task decomposition complete
  • T2: Role assignments visible

4.5. If the scenario defines a Scope Change Trigger and skip-scope-change is false:

  • Wait until Phase 2 (role assignment) is visible
  • Record T3 (scope change injection timestamp)
  • Send the scope change prompt to the team via SendMessage
  • Record T4 (scope change absorbed — role adjustment visible)

4.6. Continue observing until the team delivers its output.

  • Record T5 (integration begins)
  • Record T6 (final report delivered)

4.7. Capture the team's complete output.

Got: Team executes the task through its coordination pattern phases. Timestamps are recorded for all transitions. Scope change (if applicable) is injected and absorbed.

If fail: If the team fails to produce output, record the failure point and any error messages. If the team stalls, note the last observed phase and timeout. Proceed to evaluation with partial results.

Step 5: Evaluate Pattern Behaviors

5.1. For each key behavior from Step 3, determine whether it was observed during execution:

  • Observed: Clear evidence in the team's output or coordination
  • Partial: Some evidence but incomplete or ambiguous
  • Not observed: No evidence

5.2. For each task-specific behavior from the scenario's Expected Behaviors section, apply the same evaluation.

5.3. Record findings in the observation log.

Got: All or most pattern-specific and task-specific behaviors are observed.

If fail: Unobserved behaviors are findings, not failures of the test procedure. Record them accurately — they indicate the coordination pattern did not fully manifest.

Step 6: Evaluate Acceptance Criteria

6.1. Walk through each acceptance criterion from the scenario.

6.2. For each criterion, assign a determination:

  • PASS: Criterion clearly met with observable evidence
  • PARTIAL: Criterion partially met (counts toward threshold at 0.5 weight)
  • FAIL: Criterion not met despite opportunity
  • BLOCKED: Could not evaluate (pre-condition failure, team timeout, etc.)

6.3. If the scenario includes Ground Truth data, verify reported findings against it:

  • Calculate accuracy percentages per category
  • Flag false positives and false negatives

6.4. If the scenario includes a Scoring Rubric, score each dimension 1-5 with brief justification.

6.5. Calculate summary metrics:

  • Acceptance: X/N criteria passed (PARTIAL counts as 0.5)
  • Threshold: PASS if >= threshold defined in scenario
  • Rubric total: X/Y points (if applicable)

Got: All acceptance criteria have a determination. Summary metrics are calculated.

If fail: If fewer than half the criteria can be evaluated (too many BLOCKED), the test run is inconclusive. Document why and recommend re-running after fixing pre-conditions.

Step 7: Generate RESULT.md

7.1. Create tests/results/YYYY-MM-DD-<target>-NNN/RESULT.md using the Recording Template from the scenario's Observation Protocol.

7.2. Populate all sections:

  • Run metadata (observer, timestamps, duration)
  • Phase log with all recorded timestamps
  • Role emergence log (for adaptive/team tests)
  • Acceptance criteria results table
  • Rubric scores table (if applicable)
  • Ground truth verification table (if applicable)
  • Key observations (narrative)
  • Lessons learned

7.3. Include the team's raw output as an appendix or in a separate file (team-output.md) in the same result directory.

7.4. Add a summary verdict at the top:

**Verdict**: PASS | FAIL | INCONCLUSIVE
**Score**: X/N criteria (Y/Z rubric points)
**Duration**: Xm

Got: Complete RESULT.md with all sections populated and a clear verdict.

If fail: If result file cannot be written, output the results to stdout as a fallback. The evaluation data should never be lost.

Validation

  • Test scenario file loaded and all required sections present
  • Pre-conditions verified (or documented as BLOCKED)
  • Coordination pattern key behaviors loaded from registry
  • Team spawned and task delivered
  • Scope change injected at the right time (if applicable)
  • All pattern-specific behaviors evaluated (observed/partial/not observed)
  • All acceptance criteria have a determination (PASS/PARTIAL/FAIL/BLOCKED)
  • Ground truth verification completed (if applicable)
  • RESULT.md generated with all sections populated
  • Summary verdict calculated and recorded

Pitfalls

  • Evaluating output quality instead of coordination: This skill tests how the team coordinates, not whether the task output is perfect. A team that coordinates well but finds only 7/9 broken refs still demonstrates the pattern.
  • Injecting scope change too early: Wait until role assignment is clearly visible before injecting the scope change. Too early means the team hasn't differentiated yet, so there's nothing to adapt.
  • Conflating team member output with team output: The opaque team should present a unified output. If you see individual member reports, that's a finding about opacity, not a test infrastructure problem.
  • Exact ground truth matching: Ground truth counts are approximate. Evaluate whether findings are in the right ballpark, not whether they match exactly.
  • Forgetting to record timestamps: Timestamps are essential for measuring phase durations and adaptation speed. Set them as events happen, not retroactively.

Related Skills

  • review-codebase — deep codebase review that complements team-level testing
  • review-skill-format — validates individual skill format (this skill validates team coordination)
  • create-team — creates team definitions that this skill tests
  • evolve-team — evolves team definitions based on test findings
  • test-a2a-interop — similar testing pattern for A2A protocol conformance
  • assess-form — the morphic assessment that the opaque team lead uses internally

Dépôt GitHub

pjt222/agent-almanac
Chemin: i18n/caveman-lite/skills/test-team-coordination
0
agentsagentskillsai-assisted-developmentclaude-codeskillsteams

Compétences associées

content-collections

Méta

Cette compétence propose une configuration éprouvée en production pour Content Collections, un outil axé sur TypeScript qui transforme des fichiers Markdown/MDX en collections de données typées de manière sûre avec une validation Zod. Utilisez-la lors de la création de blogs, de sites de documentation ou d'applications Vite + React riches en contenu pour garantir la sécurité de typage et la validation automatique du contenu. Elle couvre tout, de la configuration du plugin Vite et de la compilation MDX à l'optimisation des déploiements et la validation des schémas.

Voir la compétence

polymarket

Méta

Cette compétence permet aux développeurs de créer des applications avec la plateforme de marchés prédictifs Polymarket, incluant l'intégration d'API pour le trading et les données de marché. Elle fournit également une diffusion de données en temps réel via WebSocket pour surveiller les transactions en direct et l'activité du marché. Utilisez-la pour mettre en œuvre des stratégies de trading ou pour créer des outils traitant les mises à jour de marché en direct.

Voir la compétence

creating-opencode-plugins

Méta

Cette compétence aide les développeurs à créer des plugins OpenCode qui s'interconnectent avec plus de 25 types d'événements tels que les commandes, les fichiers et les opérations LSP. Elle fournit la structure du plugin, les spécifications de l'API événementielle et les modèles d'implémentation pour les modules JavaScript/TypeScript. Utilisez-la lorsque vous avez besoin d'intercepter, de surveiller ou d'étendre le cycle de vie de l'assistant IA OpenCode avec une logique personnalisée pilotée par les événements.

Voir la compétence

sglang

Méta

SGLang est un framework de service LLM haute performance spécialisé dans la génération rapide et structurée pour les workflows JSON, regex et agentiques grâce à son cache de préfixe RadixAttention. Il offre une inférence nettement plus rapide, particulièrement pour les tâches avec des préfixes répétés, ce qui le rend idéal pour les sorties complexes et structurées ainsi que les conversations multi-tours. Choisissez SGLang plutôt que des alternatives comme vLLM lorsque vous avez besoin d'un décodage contraint ou que vous construisez des applications avec un partage étendu de préfixes.

Voir la compétence