# Fellow Schemas This file contains the full YAML/JSON schemas referenced by the Fellow skill. Keep this file alongside SKILL.md — it is the canonical source for experiment invocation contracts, cycle outputs, configuration defaults, and OKR definitions. --- ## Invocation contract Fellow is invoked only by Mentor. Invocation payload structure: ```yaml target: program_id: objective: benchmark: budget: type: wall_clock | task_count | token_budget | simulation_window value: constraints: max_variants: max_cycles: mutation_surface: [] protected_surface: [] promotion: threshold: automatic: rollback_on_regression: true runner: type: command | function | workflow entrypoint: timeout_seconds: metric_extractor: type: json | regex | function | structured_output source: selector: ``` --- ## Cycle output At completion Fellow emits: ```yaml cycle_id: target: baseline_score: best_variant_id: best_variant_score: improvement: decision: promote | no_change | abort artifacts_ref: rollback_ref: ``` --- ## Storage layout ``` {agent_root}/commons/data/ocas-fellow/ config.json experiments.jsonl decisions.jsonl requests_processed.jsonl intents.jsonl evidence.jsonl results/ {cycle_id}.json runs/ {cycle_id}/ baseline/ variant-001/ variant-002/ {agent_root}/commons/journals/ocas-fellow/ YYYY-MM-DD/ {run_id}.json ``` --- ## Default config.json ```json { "skill_id": "ocas-fellow", "skill_version": "2.4.0", "config_version": "1", "created_at": "", "updated_at": "", "defaults": { "improvement_threshold": 0.03, "max_variants_per_cycle": 20, "max_cycles_per_target": 5, "max_parallel_runs": 1, "plateau_window": 5 }, "retention": { "days": 90, "max_records": 10000 } } ``` --- ## OKRs Universal OKRs from spec-ocas-journal.md apply to all runs. ```yaml skill_okrs: - name: experiment_completion_rate metric: fraction of experiment cycles completing without abort direction: maximize target: 0.90 evaluation_window: 30_runs - name: promotion_rate metric: fraction of completed experiments producing a promotable variant direction: maximize target: 0.30 evaluation_window: 30_runs - name: baseline_stability metric: fraction of baselines completing successfully direction: maximize target: 0.99 evaluation_window: 30_runs - name: schedule_adherence metric: fraction of cron runs executing within 5 minutes of scheduled time direction: maximize target: 0.95 evaluation_window: 30_days - name: data_integrity metric: fraction of experiment records with complete, non-corrupted evidence direction: maximize target: 0.99 evaluation_window: 30_runs ``` --- ## Self-update procedure `fellow.update` pulls the latest package from the `source:` URL in this file's frontmatter. 1. Read `source:` from frontmatter → extract `{owner}/{repo}` from URL 2. Read local version from SKILL.md frontmatter `metadata.version` 3. Fetch remote version from SKILL.md frontmatter: ```bash gh api "repos/{owner}/{repo}/contents/SKILL.md" --jq '.content' | base64 -d | grep 'version:' | head -1 | sed 's/.*"$.*$".*/\1/' ``` 4. If remote version equals local version → stop silently 5. Download and install: ```bash TMPDIR=$(mktemp -d) gh api "repos/{owner}/{repo}/tarball/main" > "$TMPDIR/archive.tar.gz" mkdir "$TMPDIR/extracted" tar xzf "$TMPDIR/archive.tar.gz" -C "$TMPDIR/extracted" --strip-components=1 cp -R "$TMPDIR/extracted/"* ./ rm -rf "$TMPDIR" ``` 6. On failure → retry once. If second attempt fails, report the error and stop. 7. Output exactly: `I updated Fellow from version {old} to {new}`

ocas-fellow

Install this skill

Comments