Skip to content

hassan-mohiddin/freeflow / .deprecated/skill-tooling-pre-rewrite/skills/evaluate-skill

evaluate-skill

Verified

Use when evaluating, comparing, or making readiness claims about agent-skill behavior.

references/portable-execution.md

View on GitHub
# Portable Execution

Use one case fingerprint, one Pi-first executor, one evidence bundle, and fixed graders. Keep the concrete Pi boundary limited to capability probes, invocation construction, event parsing, usage extraction, and cleanup. Add another host only when a real supported host exists.

## Capability Rule

Choose the cheapest mode that preserves the eval question. If a fallback changes the question or evidence quality:

- mark it Diagnostic, Reduced-Fidelity, or Unsupported;
- name the missing capability;
- stop when acceptance requires it.

## Subject Isolation

For each run:

- copy only the fixture into an isolated temporary root;
- copy only the variant's declared immutable subject resources separately;
- for composition, materialize each named base/target skill and runtime resource separately, preserve order, and fingerprint every declared byte;
- keep cases, assertions, reports, controls, and labels coordinator-side;
- allow reads only from fixture and snapshot;
- allow writes only inside fixture;
- reject traversal and symlink escapes;
- expose no unrestricted shell;
- verify the snapshot hash after execution.

Directory placement alone is not isolation. If the adapter cannot enforce every exposed tool, do not claim strict isolation.

## Pi Bootstrap

Use one-shot JSON, no session, no context files, no auto-discovered resources, one explicit root guard, explicit skill snapshots, minimal tools, and an isolated config home. For declared composition, use repeated explicit skill paths plus only the exact declared evaluator-owned runtime resources; never load the installed Freeflow package. Use Pi RPC only for true fixed multi-turn evidence, currently two to four scripted user turns.
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: 471kUpdated Claude CodeCodexVerifiedInstall
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: 322kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 322kUpdated Claude CodeCodexVerifiedInstall
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 availableStars: 322kUpdated Claude CodeCodexVerifiedInstall