session-audit
Out-of-cycle, read-only audit of what a Claude Code session that ran agile-skills cost and what it shipped. It edits nothing; its output is a report.
Part of agile-skills. Needs nothing. With gh in the repo it measures shipped work (merged PRs, added lines); without it the skill falls back to tickets moved to Done and says the measure is coarser. No Jira or Confluence.
What you get
| Skill | What it does | Say |
|---|---|---|
session-audit | Spend per skill and per subagent, a before/after comparison normalised by merged PRs and added lines, a check that the gates still held, and a ranked list of skill or agent edits that would cut spend | "audit this session's cost", "compare before and after sessions", "is the new version more efficient", "where do the tokens go", "subagent cost breakdown" |
flowchart LR
t["transcripts<br/>(.jsonl, offline)"] --> u["session_usage.py<br/>cost profile"] --> r["report"]
g["gh: merged PRs,<br/>added lines"] --> r
r --> i["ranked edits<br/>to skills and agents"]
Install
/plugin marketplace add cedricfarinazzo/agile-skills
/plugin install session-audit@agile-skills
For Codex (audits supplied Claude Code transcripts only; native Codex sessions are not supported):
codex plugin marketplace add cedricfarinazzo/agile-skills
codex plugin add session-audit@agile-skills
Invoke it as /session-audit:session-audit in Claude Code or $session-audit in Codex.
Bundled script
skills/session-audit/scripts/session_usage.py reads transcripts offline and prints the usage profile. It sends nothing anywhere.
python3 session_usage.py find <title-or-session-id>
python3 session_usage.py report --run before=a.jsonl --run after=b.jsonl,c.jsonl [--agents] [--price opus=<in>,<out>]
- A run may span several transcripts; messages shared by a resumed or forked session are counted once.
- Subagent transcripts are read from
<session>/subagents/beside each file. --shipped NAME=<merged PRs>,<added lines>adds the efficiency table: cost and units per merged PR and per added line, with the gain between runs.- Default prices are assumptions; pass
--pricefor real ones.
Tests: python3 -m pytest session-audit/skills/session-audit/scripts.
Where it fits
Out of cycle. Run it after a drain, an implement pass or a merge train, or to compare two plugin versions on similar work. It does not replace the retro or the sprint closeout. Full cycle: root README.