dbc-oduffy Code Health — Night Shift Commit Review
The "night shift colleague." Queries today's completion entries to identify surfaces that saw
recorded work, dispatches a Sonnet reviewer with --problems-only, applies findings inline,
defers complex findings to the debt backlog, updates the health ledger, writes a morning-ready
summary. Results wait at the next workstream-start.
Announce at start: "I'm using /code-health to review recent commits."
Run on every committed day, regardless of commit count. A small commit count is exactly when a review gets skipped and is exactly where the adjacent-path regression hides — the cost-benefit is asymmetric. The only valid skip: zero completion entries today AND zero fallback commits (see Failure Modes).
Step 1: Identify Surfaces
Invoke through the .exe launcher by absolute path via the PowerShell call operator (Shape W).
Ladder and shapes: snippets/resolve-coordinator-bin.md.
`& "$env:COORDINATOR_SETTINGS_HOME\bin\query-completions.exe" --where "created=<YYYY-MM-DD>" --format json`
Extract file paths / subsystem names from title, description, files. No entries for today:
read state/health-ledger.md's Last daily check: date, fall back to
git log --since="<last-check-date>" --oneline --stat, update the timestamp, report the fallback,
continue with the commit-based scope.
Step 2: Diff Scope
git diff HEAD -- <file1> <file2> ... over Step 1's surfaces (a directory prefix if Step 1
yielded a subsystem name; git diff <last-check-commit>..HEAD on the fallback path). Summarize
which files/systems changed — this drives the reviewer's vocabulary/emphasis in Step 3, not
reviewer selection.
Step 3: Dispatch the Reviewer
Always the Sonnet coordinator:code-reviewer, never a persona — personas are Opus-only, reserved
for the weekly arch pass and explicit architectural decisions; a finding that genuinely needs one
gets flagged for /workweek-complete Step 7.5, not escalated here. Tell it what to weight by
dominant change type (UE idioms / component-token-reuse-accessibility / numeric-correctness /
coupling-and-interface-seams) — vocabulary only, never identity.
Dispatch unattended: coordinator:code-reviewer, UNNAMED, run_in_background: true,
--problems-only. Its sidecar is spawn-provisioned (arrives as sidecar_path: in its brief) — no
EM pre-scaffold. Read the returned DONE: <sidecar-path> | verdict: <OK|WARN|BLOCKED> | findings: <N> and pass the path to Step 4.
Step 4: Apply Findings
Dispatch coordinator:review-integrator on the on-disk sidecar path (never inline-relayed
findings) plus affected file paths. It applies inline fixes and annotations. Findings needing 3+
interacting files or new abstractions go to Step 5 instead. No findings: skip to Step 6.
Step 5: Debt Backlog
`& "$env:COORDINATOR_SETTINGS_HOME\bin\coordinator-queue-append.exe" --schema debt-backlog`
One YAML file per finding under state/debt-backlog/. Required: title, body (block scalar:
observation, structural gap, context), source: daily-health/code-reviewer/{date}, risk,
proposed_action, status: open, created. Stage each: git add state/debt-backlog/<date>-<slug>.yaml.
Step 6: Health Ledger
Create state/health-ledger.md from template if absent (header Last daily check:/Last full audit:/Next rotation target:, then a ## System Index table: System | Grade | Status | Last
Audited | Open P0/P1/P2 | Lines | Notes). Update Last daily check to today. If findings changed
a grade, update that row; a system touched but rowless gets grade ?.
The health ledger is the single source of truth for grades — /architecture-audit also writes
here after weekly audits. Read the existing grade before changing it; don't downgrade a
just-upgraded system absent new P0/P1s.
Grading anchors, status-trigger definitions, and the reviewer-emphasis-by-change-type mapping this step and Step 3 apply by eye have no engine producer yet — apply the calibration in the wiki page for this skill until one lands.
Step 7: Health Summary
`& "$env:COORDINATOR_SETTINGS_HOME\bin\coordinator-doc-new.exe" --type health-status --title "Health Summary"`
Via Edit, fill health: (HEALTHY|WATCH|ACTION|CRITICAL), summary: (one-line), and body
sections ## Systems Graded, ## Findings Summary, ## Action Items for Next Session. Worked
example: wiki.
Step 8: Commit
Commit per snippets/scoped-commit-route.md, subject daily-code-health: review of surfaces from completion entries [date], pathspec exactly state/health-ledger.md and
state/health/<YYYY-MM-DD>-health-summary.md.
Nothing else this run touched. Post-commit hook pushes automatically.
Failure Modes
| Situation | Action |
|---|---|
| No health ledger on first run | Create from template, use last 24h as scope |
| No completion entries for today | Fall back to git log --since=<last-check> scope; report fallback |
query-completions not found |
Fall back to git-log scope; note missing binary |
| No new commits since last check (fallback path) | Update timestamp, report, exit — no dispatch |
| Reviewer returns no findings | Skip Steps 4-5, go to Step 6 |
| Debt backlog doesn't exist | Create from template first |
| Complex finding can't be fixed inline | Debt backlog, with severity and effort estimate |
| Git commands fail (no commits, detached HEAD) | Report the error and stop |
| review-integrator unavailable | Log findings into the health summary body manually, note as deferred |
Cost
1 Sonnet code-reviewer dispatch (--problems-only) + 1 review-integrator dispatch if findings
exist. No persona, no Opus at this cadence.
Relationship to Other Commands
/workday-complete is the primary trigger — let it invoke this rather than running standalone.
/workstream-start's cockpit snapshot and record queries surface state/health/*.md at the top
of the next session (nested path, not the stale flat state/health-summary.md). /review-code's
full feature-review workflow is a different tool — this dispatches --problems-only directly.
pipelines/daily-code-health/PIPELINE.md is the pipeline definition this command executes.