Checklists

A checklist is the contract between IonWarp and you about what "working" means. Every project carries two: the onboarding list ("Ship your first review") and Beast Mode, the daily operations bar that grades every review lens on one standard. Item states are derived on every read — a green check means real telemetry said so, never that an agent claimed it. Beast Mode is evaluated once a day by a scheduled beast-mode-checklist run that publishes the day's verdicts, evidence, and a one-line summary; the same rows feed the card on the home page, this API, and the bottom of the weekly report.

Read a project's checklists

curl -H "Authorization: Bearer $DASH_PAT" \
  "https://ionwarp.com/api/v1/projects/<project_id>/checklists?include_history=7"

The response returns every list with derived item states. include_history=7 adds Beast Mode day rows, each {day, summary, verdicts[], cost_usd} — one row per UTC day, newest first, where each verdict is {item_id, verdict: "pass"|"fail"|"skip", detail}.

Freshness — green, yellow, red

Beast Mode's freshness is derived at read time and never stored: the latest published day equal to today (UTC) is green, yesterday is yellow, anything older — or no run at all — is red. No review lens running today means red; success is never inferred from silence.

The HSF-E standard

Every lens (security, code review, test holes, complexity, plan review, signal/noise grading) is graded on ONE number set:

• HSF — delivered findings with severity critical or major (the verified high-signal proxy; there is no separate "verified" flag).
• reviewed_mb — megabytes of the embedded diff actually reviewed (review_scope.reviewed_bytes), not the full gh pr diff.
• usd_per_mb = cost_usd / reviewed_mb · hsf_per_usd = hsf_count / cost_usd · hsf_e = hsf_per_usd / usd_per_mb.

Worked example: a security lens that spent $1.10 on 0.4 MB of diff and delivered 2 critical/major findings scores usd_per_mb = 2.75, hsf_per_usd = 1.82, hsf_e = 0.66 — a pass (the default floor is hsf_per_usd ≥ 0.5). A lens with zero findings still passes when usd_per_mb ≤ 10: a clean PR is not inefficiency when scan cost was proportionate to scope.

Update from an agent (MCP)

• checklist_list — the evaluated state (pass include_history: 7 for the day rows).
• checklist_item_update — an agent may write in_progress or blocked (with a named gap) plus detail/href/evidence. done is refused (checklist_done_is_derived).
• checklist_run_publish — Beast Mode's ONE terminal write: the day's verdicts and summary. Every declared judgment item must be verdicted or the whole publish fails (checklist_verdict_missing); publishing again the same UTC day replaces that day's record.

Let the agent do it

The beast-mode-checklist skill runs once a day per project on its own schedule (visible under Settings → Usage → Scheduled Tasks, with run-now, pause, and resume). It reads the day's review telemetry and per-lens spend, computes each lens's HSF-E verdict, and ends in exactly one checklist_run_publish.