Evolve Skill From Traces
Skill Verifiziert AktivEvolve SKILL.md files from agent execution traces using a three-stage pipeline: trajectory collection from observed runs, parallel multi-agent patch proposal for error and success analysis, and conflict-free consolidation of overlapping edits via prevalence-weighting. Based on the Trace2Skill methodology.
To automate the creation and refinement of SKILL.md files by learning from observed agent execution traces, ensuring that documented procedures accurately reflect effective agent behavior.
Funktionen
- Trajectory collection from agent execution logs
- Parallel multi-agent patch proposal for error and success analysis
- Conflict-free consolidation of edits via prevalence-weighting
- Validation against held-out traces for skill accuracy
- Automatic registration or version-bumping of evolved skills
Anwendungsfälle
- Evolving existing SKILL.md files when observed agent behavior differs
- Generating new SKILL.md files from scratch using recorded expert agent demonstrations
- Improving skill robustness by analyzing failure modes from execution traces
- Consolidating conflicting improvements proposed by multiple agents
Nicht-Ziele
- Manually writing SKILL.md files from scratch without trace data
- Directly executing agent code; focuses on skill file generation
- Replacing human oversight entirely; provides a framework for automated evolution
Practical Utility
- info:Usage examplesThe SKILL.md provides example bash commands for trace partitioning and validation, but lacks end-to-end examples for the entire pipeline.
Installation
/plugin install agent-almanac@pjt222-agent-almanacQualitätspunktzahl
VerifiziertVertrauenssignale
Ähnliche Erweiterungen
External Context
100Invoke parallel document-specialist agents for external web searches and documentation lookup
Swarm Orchestration
100Orchestrate multi-agent swarms with agentic-flow for parallel task execution, dynamic topology, and intelligent coordination. Use when scaling beyond single agents, implementing complex workflows, or building distributed AI systems.
Evolve Agent
99Evolve an existing agent definition by refining its persona in-place or creating an advanced variant. Covers assessing the current agent against best practices, gathering evolution requirements, choosing scope (refinement vs. variant), applying changes to skills, tools, capabilities, and limitations, updating version metadata, and synchronizing the registry and cross-references. Use when an agent's skills list is outdated, user feedback reveals capability gaps, tool requirements have changed, an advanced variant is needed alongside the original, or the agent's scope needs sharpening after real-world use.
Create Skill
99Create a new SKILL.md file following the Agent Skills open standard (agentskills.io). Covers frontmatter schema, section structure, writing effective procedures with Expected/On failure pairs, validation checklists, cross-referencing, and registry integration. Use when codifying a repeatable procedure for agents, adding a new capability to the skills library, converting a guide or runbook into agent-consumable format, or standardizing a workflow across projects or teams.
Create Agent
99Create a new agent definition file following the agent-almanac agent template and registry conventions. Covers persona design, tool selection, skill assignment, model choice, frontmatter schema, required sections, registry integration, and discovery symlink verification. Use when adding a new specialized agent to the library, defining a persona for a Claude Code subagent, or creating a domain-specific assistant with curated skills and tools.
Omc Plan
99Strategic planning with optional interview workflow