Agent KPIs
This guide defines recurring metrics for agent-focused product and marketing execution.
Core KPIs
agent_e2e_success_ratetime_to_green_secedits_countfailures_recovered
Proof KPIs
- Claim reproducibility coverage (claims with command + artifact + date).
- Proof artifact freshness age in days.
- Weekly proof report completion (
composer proof-weekly-reportgenerated and reviewed).
Weekly Funnel Capture Fields
week_start(ISO date)repo_visitsclone_countstars_deltaissues_agent_regression_reportissues_agent_workflow_requestissues_agent_proof_requestproof_commands_run_count(manual log until automation is added)
Cadence
- Weekly: refresh proofs and check freshness.
- Monthly: regenerate benchmark report and baseline case-study payload.
- Monthly: review KPI movement against distribution actions and adjust posting mix.
Commands
composer proof-refresh
composer proof-freshness-check
composer proof-weekly-report
composer marketing-weekly-cycle
Weekly Automation
Use one command to run the weekly operational loop:
composer marketing-weekly-cycle -- \
--repo-visits=0 \
--clone-count=0 \
--stars-delta=0 \
--issues-agent-regression-report=0 \
--issues-agent-workflow-request=0 \
--issues-agent-proof-request=0 \
--proof-commands-run-count=4 \
--notes="Update with real GitHub traffic numbers"
Outputs:
docs/guides/proof-weekly-YYYY-MM-DD.md- updates row for the week in
docs/guides/Agent-KPI-Log.md docs/guides/distribution-proof-drop-YYYY-MM-DD.md
Optional GitHub API Ingestion
When a token is available, the same command can auto-ingest:
repo_visitsandclone_count(traffic endpoints)- issue counts for
agent+bug,agent+enhancement,agent+documentation stars_delta(weekly stargazer delta)
GITHUB_TOKEN=... composer marketing-weekly-cycle -- \
--github-fetch=1 \
--github-owner=coesion \
--github-repo=core \
--notes="Weekly KPI refresh from GitHub API"
Notes:
- Explicit numeric flags (for example
--repo-visits=...) override fetched values. - If GitHub endpoints are unavailable, the script keeps existing/manual values and logs a warning.
Monthly Review Checklist
- Compare monthly totals for visits, clones, stars, and agent-template issue counts.
- Identify which proof posts produced visible KPI movement.
- Remove or rewrite low-performing distribution copy/templates.
- Confirm all canonical claim links still point to
docs/AUDIT.mdproof rows. - Define one experiment for next month (audience/channel/message) and track its impact.