lanyasheng/auto-improvement-orchestrator-skill
13 skills13 installs78 starsGitHub
Install
npx skills add https://github.com/lanyasheng/auto-improvement-orchestrator-skillSkills in this repo
1Autoloop ControllerRuns the 5-stage improvement pipeline repeatedly with plateau/oscillation convergence detection, budget control, and disk-persisted state. A developer uses it for unattended, multi-round automated improvement of a skill until quality converges.1installs2Benchmark StoreMaintains a benchmark database of frozen tests, quality tiers, and a Pareto front for the skill-improvement pipeline. A developer uses it to baseline scores, detect regressions, and feed the improvement gate.1installs3Improvement DiscriminatorA multi-signal scoring engine that ranks improvement candidates via heuristic rules, LLM-as-Judge semantic evaluation, and blind multi-reviewer panels. A developer uses it in the pipeline's second stage to prioritize candidates for execution.1installs4Improvement EvaluatorMeasures whether a skill change actually improves AI task performance by running a predefined task suite and reporting an execution pass rate. A developer uses it to validate improvements with concrete execution metrics for gating decisions.1installs5Improvement ExecutorApplies accepted improvement candidates to files using four action types with automatic backups and a rollback pointer. A developer uses it in the pipeline's fourth stage to safely apply or revert skill file changes.1installs6Improvement GateValidates executed improvement candidates through six gate layers, with schema/compile/regression/review blocking and lint/human-review advisory. A developer uses it in the pipeline's final stage to keep, reject, or queue changes for human approval.1installs7Improvement GeneratorProduces ranked improvement candidates for a target skill from target analysis, feedback signals, and failure traces. A developer uses it as stage 1 of the auto-improvement pipeline, or standalone to propose fixes for one skill.1installs8Improvement LearnerScores a skill's structural quality on 9 dimensions and runs an evaluate-modify-re-evaluate self-improvement loop that keeps or reverts changes. A developer uses it to diagnose why a skill scores low and to auto-optimize SKILL.md structure.1installs9Improvement OrchestratorCoordinates the five-stage improvement pipeline (Generator, Discriminator, Evaluator, Executor, Gate) for one or more skills. A developer uses it to run a complete improvement cycle with trace-aware retries rather than invoking each stage manually.1installs10Prompt HardeningProvides a systematic method and audit script for hardening prompts so agents stop ignoring or creatively circumventing rules. A developer uses it when an agent repeatedly violates the same rule or before deploying a new system prompt.1installs11Session Feedback AnalyzerMines Claude Code session logs for implicit feedback signals and outputs feedback.jsonl with per-skill correction rates and dimension hotspots. A developer uses it to find which skills users correct most and to feed prioritization into the improvement pipeline.1installs12Skill DistillConsolidates N functionally overlapping skills into a single distilled SKILL.md plus references, resolving trigger conflicts and token bloat. A developer uses it when several skills in one domain overlap and users cannot tell which to load.1installs13Skill ForgeGenerates evaluation task suites from a SKILL.md by extracting testable claims from five sections and assigning judge types, or scaffolds a new skill from a spec. A developer uses it to add runnable tests to a skill or generate one from requirements.1installs