auto-skill-build-agent-run-ledger-dashboard

Append and summarize a lightweight local `.agent-runs/runs.jsonl` ledger (claudetop-inspired) so Codex sessions build durable observability over time. Use at the end of a significant task, after runni

SKILL.md

GitHub

Agent Run Ledger Dashboard

Generated by: Codex Supercharge maintenance automation.

Goal: cheap, append-only run tracking without relying on proprietary telemetry.

Append a run entry

Run from the repo you worked in:

python3 plugins/codex-supercharge/scripts/agent_runs_append.py \
  --task "Fix flaky auth tests" \
  --model "${CODEX_MODEL:-}" \
  --status ok \
  --command "pytest -q" \
  --notes "Root cause: clock mocking"

Notes:

  • --model defaults to $CODEX_MODEL when set.
  • The script records current git branch/head + uncommitted diff stats when available.

View a dashboard summary

python3 plugins/codex-supercharge/scripts/agent_runs_report.py --days 7 --last 20

Recommended workflow

  1. After completing a task, run the append command once.
  2. When starting work, run the report command to see recent context.
  3. If you want richer fields (tokens/cost), add them as new keys; keep JSONL append-only.

Skip when

  • The repo is sensitive and you cannot persist local logs.
  • The task is a trivial one-liner and you are not doing longitudinal tracking.

References

  • plugins/codex-supercharge/references/agent-run-ledger.md
  • Pattern inspiration only (do not run blindly): sources/claudetop

Validation

  • Appending creates exactly one JSONL record for the run.
  • Reports summarize the requested window without rewriting the ledger.
  • Sensitive task notes are summarized or omitted before persistence.