Automate goal-driven development cycles by building, evaluating, and iterating on features.
npx skills add https://github.com/yuyuxinli/moodcoco --skill evolve-yuyuxinli
Please help me install this Agent Skill. Skill: evolve Source: https://github.com/yuyuxinli/moodcoco/tree/main/.claude/skills/evolve Command: npx skills add https://github.com/yuyuxinli/moodcoco --skill evolve-yuyuxinli
Help teams and agents turn an ambiguous goal into an executable program and then autonomously build, evaluate, and iterate on features until the goal's evaluation thresholds are met, reducing manual orchestration and follow-up.
Start an autonomous improvement run by issuing '/evolve' to initialize and then run '/loop 1m /evolve' to let the agents auto-build and evaluate repeatedly.
High-intent search queries and answers about installing and using this skill.
To automate build-evaluate loops, initialize with '/evolve' to generate a program spec and adapter, then run '/loop 1m /evolve' to let Builder and Critic agents auto-build, run tests, and iterate until evaluation thresholds are met.
An autonomous build-evaluate loop alternates Builder and Critic agents to implement features, run deterministic checks, and call independent LLM evaluators. It records structured progress in .evolve/results.tsv and iterates until acceptance criteria are met.
Yes, you need Git and Python 3.8+ to run autonomous agent testing loops. You also need a project-specific adapter implementing setup, run_checks, and teardown methods, plus an independent evaluator CLI such as codex or claude.
Autonomous agents prevent runaway changes using concurrency locks, stop conditions, trajectory analysis, and strict file-write rules like append-only logs and read-only program contracts. This ensures safe recovery and controlled iterations.
Yes, you can use independent LLM evaluator CLIs such as codex or claude to evaluate code iterations. The autonomous loop calls these external evaluators to assess features against goals and record structured progress in .evolve/results.tsv.
The best way is to initialize with '/evolve' to guide goal definition, generate a testable program.md spec, and create a project adapter. Then run '/loop 1m /evolve' to let agents autonomously build, evaluate, and iterate until acceptance criteria are met.
giosuetedeschi-specCreate isolated Git worktrees with safety checks and automated environment setup.
ritual-foundationDetect agent trajectory divergence and trigger circuit-breaker actions.
ether-moonRun parallel LLM CLI reviews and synthesize prioritized reports.
arthtyagiProvision Vercel Sandboxes to run benchmark scenarios and generate coverage reports.
