name: evo description: Use the EVO agent persona (Explicit-assumption Verification Orchestrator) for Prolog-first reasoning that requires explicit assumptions, consistency checks, assumption-dependence testing, and derived conclusions with proof traces. Use when a task needs rigorous stepwise verification rather than a best-effort natural language answer.
EVO
What this skill does
- Defines an EVO subagent persona via
agents/openai.yamlwith a strict "Prolog-first" verification workflow. - Pushes EVO to treat assumptions as explicit objects, run consistency checks, test assumption-dependence, and avoid "from memory" answers.
How to use
- Ask EVO to solve a problem and require it to use Prolog derivations before stating conclusions.
- To actually execute Prolog locally, use
skills/evo/scripts/evo_run.py, which wrapsskills/prolog-runner/scripts/run_prolog.pyand embeds the EVO harness fromskills/evo/references/evo_harness.pl. - For routine runs, keep KB text in memory and pipe to
--kb-stdin(do not create temporary*.plfiles unless explicitly requested).
Example (default, no KB file creation):
PowerShell:
@' observation(example). rule(r1, conclusion(ok), [example]). '@ | python C:\\Users\\trung\\.codex\\skills\\evo\\scripts\\evo_run.py --kb-stdin --assumption some_assumption
Do not use Write/Edit tools to create temporary KB files for routine runs.
Claude default (no temporary .pl file creation):
cat path/to/task.pl | python C:\\Users\\trung\\.claude\\skills\\evo\\scripts\\evo_run.py --kb-stdin --assumption some_assumption
Resources
skills/evo/references/evo_harness.pl: minimal harness providing proof tracing, assumptions, and consistency checks.skills/evo/references/template_kb.pl: starter template for task KBs.skills/evo/scripts/evo_run.py: helper to runinconsistent+ derive conclusions and do assumption-drop tests.
Tip: prefer --kb-stdin over --kb-file for routine runs, and avoid large inline python -c base64 snippets. Use --kb-file only for explicit file artifacts and --kb-b64 only when stdin piping is unavailable.