richfrem/agent-plugins-skills / plugins/agent-orchestration/.apm/skills/evaluating-agent-outcomes
evaluating-agent-outcomes
VerifiedRepository-independent evaluation of delegated agent work: assesses a delivery against a work contract (criterion-level pass/fail/unverified, failure conditions, budget, authorization), checks evidence validity, selects proportionate verification, measures cost with missing data kept unknown, classifies human interventions, and records and retrieves specific lessons. Use when orchestrating-agent-work needs findings after a worker returns, or to gate a benchmark before model calls. Not an orchestrator and never grants authority.
Evaluating Agent Outcomes (evaluating-agent-outcomes)
The evaluator supplies findings; the orchestrator decides. Scores and passing checks never expand authorization.
Contents
Constraints
- Findings only:
completeis true only when every required criterion passes and no failure condition fired. A numeric score cannot change it. - Unknown is not zero: missing usage, interventions or telemetry are reported as unknown with coverage.
- Bounded storage: records go to an explicit store inside a declared root (default
<root>/.agent-work/records.jsonl); never to a global ledger. - One implementation: reusable logic lives in
scripts/(eval_core.pygates and records,scoring.py,accounting.py,usage_measure.py,outcome_eval.py,lessons.py); consumers supply profiles, policy and storage and import it (Sanctuary does). Do not re-implement scoring or accounting in a consumer. - Scope: this skill evaluates ordinary work. Project policy (research verdicts, governance, budgets, dashboards) belongs to the consuming repository's own skill.
Quick Start
python3 scripts/outcome_eval.py validate contract.json
python3 scripts/outcome_eval.py assess contract.json --root . [--evidence evidence.json] # exit 0 only if complete
python3 scripts/lessons.py --root . find "restricted PATH worker launch" # before dispatch
python3 scripts/lessons.py --root . add --trigger ... --failure ... --evidence ... --action ... --verification ... --limits ...
python3 scripts/outcome_eval.py assess contract.json # optional rubric scores via scoring.py (assets/rubrics/pr-review.json)
python3 scripts/eval_cli.py --root . validity benchmark.json # include a "claim" object for claim-specific validity # fixtures | pilot | validity | intervention | repair | invest | summary
Workflow
- Before dispatch:
lessons.py find. 2. After a worker returns:outcome_eval.py assessand hand findings to the orchestrator. 3. Before model benchmarks:eval_cli.py fixtures|pilot; after:validity. 4. Recordintervention/repair/lessons.py add|usefrom observed events.
Verification
python3 -m pytest tests -q
Defaults
Small by default: acceptance, evidence, resource use, blockers. A rubric is optional (assets/rubrics/pr-review.json is an example profile). Deterministic checks first; bounded model review only where assertions cannot settle the question.
References
- work-contract.md — contract, findings and record formats, storage rules.
- scoring-and-accounting.md — Profile-driven scoring, cost tiers, usage ledger folding, usage measurement; what consumers keep.
- gates.md — fixtures, output-contract pilot, benchmark validity cap, interventions, repair recurrence, investment accounting.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 469k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 321k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 321k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 321k | Updated | Claude CodeCodex | Verified | Install |