Audit a launch at one stage: preparation, execution, or outcomes. Review evidence for that stage, identify missing information and critical failures, and receive a verdict without combining planned work, observed operations, and actual results into one score.
Use when the user asks to "audit our launch plan", "are we ready to launch", or evaluate launch execution/outcomes; runs one typed RAMP preflight, execution, or outcome profile without mixing time horizons. Not for recording launch state — use launch-registry; not for running launch day — use launch-day-conductor. 发布就绪审计/RAMP分阶段评估/发布前放行
Run score Not tested
See how to installReview the instructions the agent will receive before you install.
Audit one launch at one lifecycle read. Preflight evaluates readiness/assets plus planned policy and instrumentation red lines; execution evaluates observed launch-window operation; outcome evaluates post-lag proof. There is no cross-time composite.
Run RAMP preflight for launch alpha against the registry stage, canon, claims, rules, and event QA.
Run the outcome profile at day 30; keep it separate from the preflight result.
Reads: one launch, one lifecycle read, registry/canon/claims state, and profile-specific evidence. Writes: only a permissioned v3 artifact. Done when: the selected profile is complete or its exact Unknowns are reported, with no launch execution or registry mutation.
launch-registry owns stage/date/embargo facts. launch-day-conductor executes the runbook. This auditor only judges the frozen evidence.
| Need | Preferred evidence |
|---|---|
| Stage/access | Projected launch record plus direct access/eligibility check |
| Narrative/claims | Canon version, claims projection, rendered assets |
| Operations/rules | Launch plan, commitments, dated official platform rules |
| Instrumentation | Verified events/UTMs and destination truth checks |
| Execution | Timestamped action/incident/response evidence |
| Outcomes | Own analytics/CRM/store truth after declared lag |
../../../references/auditor-runbook.md../../../references/scoring-semantics.md../../../references/ramp-benchmark.md../../../references/runtime-invocation.mdreferences/auditor-runtime.mdRead ../../../references/auditor-runbook.md, scoring-semantics.md, ramp-benchmark.md, and the RAMP catalog entry. Standalone installs use bundled immutable references/auditor-runtime.md; never fetch mutable main. Before deterministic calls, follow runtime-invocation.md, resolve AARON_SKILLS_ROOT="${CLAUDE_PLUGIN_ROOT:-$(git rev-parse --show-toplevel 2>/dev/null || true)}", and require the scorer, validator, and typed catalogs. If unavailable, return score_state: NOT_SCORED / score_confidence: not_scored with no gate verdict or persistent artifact.
Declare profile/lifecycle read (preflight|execution|outcome), target launch, launch type, market, access model, observation date, and evidence window.
Every observed state needs source/date/type/confidence. Missing applicable evidence is Unknown; catalog-authorized conditional items may be N/A with reason. Run python3 "$AARON_SKILLS_ROOT/scripts/rubric-score.py" score <run.json> on the selected typed profile.
Verify profile-relevant vetoes: RAMP-R1 stage/access contradiction, RAMP-A1 material claim/disclosure failure, RAMP-M1 planned or observed manipulation/embargo/platform violation, and RAMP-P1 demonstrably broken instrumentation on participating surfaces.
DONE/SHIP, final 80.DONE_WITH_CONCERNS/FIX, final 59.DONE/BLOCK, no final score.NEEDS_INPUT/UNDECIDED, no score.State lifecycle read with every result. On trace request, qualify RAMP-R1/A1/M1/P1, especially against colliding ROAS IDs.
Begin with the auditor-runbook's exact typed conversation header. Never replace status, verdict, or score_state with prose; list each explicitly missing qualified item as ``ID: unknown``` before findings. Use the stable catalog ID verbatim (for example, RAMP-R1); never substitute an evidence-subcheck label or synthesize a suffixed ID such as RAMP-R1d`.
Lead with lifecycle-specific verdict, target/context/date, score or coverage/interval, confidence, profile detail, linked prior reads, critical evidence, Unknowns, and fix/rerun owner. A preflight SHIP authorizes no external launch action by itself; explicit execution approval remains required.
Persist only after explicit authorization to memory/audits/launch/YYYY-MM-DD-<topic>.md. Preserve the scorer's orthogonal status and verdict; validate the complete v3 draft with validate-audit-artifact.py against the intended relative path, persist only through one full-content Write, and revalidate the target per the auditor runbook. Edit/shell/MCP mutations of the reserved sink are unsupported. Create separate files for separate lifecycle reads and link them by launch ID rather than overwriting.
No reviews yet. Tell others what you thought.
No login needed. Share how it worked for you, ask a question, or note where it did not run.
No comments yet. Be the first to write one.
Audit paid advertising accounts before increasing spend by comparing campaign exports with your own sales or lead records. Review 20 criteria covering measurement, spending, and safety, then receive a decision and prioritized fixes.
Create paid-ad headlines, descriptions, and opening hooks that match the destination page. Build a matrix of messaging angles and variants, check format limits, and flag unsupported claims or policy risks before handing the set off for review.
Plan advertising and landing page experiments with a clear hypothesis, controlled variants, sample sizes, and duration. Interpret exported results by separating statistical evidence from meaningful business impact, applying an action rule only when its owner approved it in advance.
Design employee and founder sharing programs with voluntary participation, personalized share kits, disclosure lines, and staggered posting windows. Get roster and Slack or Teams distribution specifications, plus safeguards against identical coordinated posts and organized engagement exchanges.
Reconcile paid advertising conversions with GA4 or ecommerce order records. Identify duplicate and unmatched claims, align attribution windows and currencies, compare how models assign credit, and assess additional orders when test and control data are available.
Turn customer interviews and won or lost deal notes into a map of buyer beliefs, recurring objections, and reasons to switch or stay. Each finding includes its evidence, with suggested ways to address objections and clear labels for assumptions.