skills/ that the agent reads and follows at runtime. Skills are the “K” in the AVO equation Vary(P_t) = Agent(P_t, K, f).
How skills work
Each skill is a.md file with:
- YAML frontmatter:
nameanddescription(first sentence = when to use) - Free-form body: instructions the agent follows
Pipeline skills
| Skill | Purpose |
|---|---|
pipeline-flow | Orchestrates the 7-phase pipeline |
research-mode | Choose speed/balanced/deep from user context |
decompose-task | Break complex questions into sub-questions |
gene-query | Generate diverse queries from entity/pain/object genes |
select-channels | Pick 5-10 most relevant channels per topic |
Knowledge skills
| Skill | Purpose |
|---|---|
systematic-recall | Extract everything Claude knows before searching |
use-own-knowledge | Contribute training knowledge as a source |
knowledge-map | Save/load structured knowledge maps across sessions |
extract-knowledge | Pull reusable knowledge from high-quality results |
Evaluation skills
| Skill | Purpose |
|---|---|
llm-evaluate | Semantic relevance judgment on search results |
rerank-evidence | Rank results by importance for synthesis |
normalize-results | Standardize results into canonical schema |
extract-dates | Extract publication dates from all available signals |
generate-rubrics | Define binary rubrics for quality evaluation |
check-rubrics | Evaluate each rubric as pass/fail with evidence |
Synthesis skills
| Skill | Purpose |
|---|---|
assemble-context | Select, deduplicate, compress evidence for synthesis |
synthesize-knowledge | Turn evidence into conceptual frameworks |
evaluate-delivery | Self-check delivery quality before presenting |
Evolution skills
| Skill | Purpose |
|---|---|
auto-evolve | Run one AVO evolution step |
anti-cheat | Reject novelty collapse and score gaming |
outcome-tracker | Learn which queries produced used results |
provider-health | Track platform health and skip cooling-down sources |
Meta-skills (immutable)
These five skills cannot be modified by evolution:| Skill | Purpose |
|---|---|
create-skill | Create new skills when a capability is missing |
observe-user | Learn user preferences and project context |
extract-knowledge | Pull reusable knowledge from results |
interact-user | Handle user dialogue and feedback |
discover-environment | Scan available tools, models, and API keys |
Utility skills
| Skill | Purpose |
|---|---|
fetch-webpage | Full content extraction from URLs |
follow-links | Follow outlinks from curated collections |
search-author-track | Discover a researcher’s full body of work |
search-citation-graph | Follow citation chains from known papers |