Autonomous Contribution

Autonomous Contribution Skill

CONCEPT:AU-AHE.evaluation.backtest-harness Distributed Agentic Evolution

This skill orchestrates the packaging and submission of locally evolved intelligence (new SKILL.md files or highly successful TeamConfig nodes) back to the central agent-packages repository.

Triggers

This skill is invoked automatically by the genius-agent --evolve background daemon when a local SelfImprovementCycle yields a new artifact that passes the local verification threshold.

Required Telemetry

When preparing a Pull Request, you MUST ensure that the payload contains the mandatory AU-ECO.ui.company-infrastructure-orchestration Community Telemetry:

1. origin: Set to "community".
2. timestamp: The precise ISO-8601 timestamp of when the artifact was verified.
3. author: The deterministic origin hash of the agent that generated the artifact.
4. Guardrail: All new skills must include Author: Autonomous in the frontmatter of the generated SKILL.md.

Workflow

1. Create the artifact: serialize the TeamConfigNode to JSON or format the CallableResourceNode metadata into a standard SKILL.md package, embedding the telemetry fields above.
2. Start a new branch prefixed with evolve/ (e.g., evolve/team-config-12345 or evolve/skill-new-feature) using git_tools.
3. Add the files and commit with a standard semantic commit message (e.g., feat(evolution): add autonomous skill <name>).
4. Use github-tools to push the branch to the remote origin and open a Pull Request against the main branch. The PR body must clearly explain the performance metrics (e.g., composite_score) that justified the promotion.

Use the skill directly to package and open a single artifact's PR. Delegate a batch of several evolved artifacts so each is branched, committed, and opened independently, keeping every submission behind the same human-review boundary.

Human-in-the-Loop

Do not attempt to auto-merge the Pull Request. The central repository requires a human maintainer to review and approve all autonomous contributions before they are ingested globally via engine_ingestion.py. Use an economy model for the routine packaging and formatting steps above; escalate to a stronger model only when the performance metrics justifying the promotion are ambiguous.