fidelity
Use when the user runs /fidelity. Computes the plan-fidelity rate — what fraction of /mandos verdicts came back Faithful, and the Missing-vs-Scope-crept Blocker split that explains the rest — from the
SKILL.mdskills/fidelity/SKILL.md
GitHubFidelity — The Plan-Fidelity Rate
Where /este asks whether the config's rules and skills were followed, /fidelity asks whether the plans they governed turned into matching code. Every /mandos verdict — Covered / Missing / Scope-crept — is recorded as it renders; this skill reduces that record into one standing rate, split by which way the drift ran.
Run
python3 ~/.claude/hooks/fidelity-scan.py
This reads ~/.skadi/mandos/history.jsonl (read-only — written by mandos-record.sh as a side effect of /mandos), writes the board channel ~/.skadi/board/fidelity.json, and renders ~/.skadi/henneth/plan-fidelity.html. It prints the headline Faithful rate.
Then ensure the Henneth window is up (boot it with /henneth if not) and surface the dashboard URL inline — the page follows the folder, so the dashboard appears on its own once rendered.
Notes
- Read-only. The scan reads the recorded history; it never writes to a tracker or repo.
- Sourced from
/mandos, not transcripts. Unlike/este's marker scan, this rate has no signal until/mandoshas actually been run a few times — an unused history readstotal: 0and the dashboard says so plainly, rather than guessing. - Blocker split localizes the leak.
missingBlockers(execution came up short of the plan) andscopeBlockers(execution went beyond the plan) are counted separately — only Blocker-severity items gate a/mandostier, so they are what actually explains why the rate isn't higher. - Standing cadence. The skill arms nothing. For a recurring read, wrap it:
/loop 6h /fidelity, or a cron. - Tests ride beside the hook —
fidelity-scan.test.shruns offline via injected fixture history (MANDOS_HISTORY_DIR).