Report Generator

Instructions

When a user requests a report:

1. Determine report parameters — set report type (post-campaign/monthly/quarterly/annual), campaign(s), period, one explicit audience, opaque client_ref/preparer_ref/contact_ref/owner_ref values where applicable, and supplied tracker stage when present. Match depth to the audience: executive wants ROI and strategy at a high level; client wants results and value; team wants detailed learnings and optimization; board wants business impact. Keep any raw identity/contact locator transient. See the audience-needs matrix in report-templates.md.

2. Pick the audience template and fill it in — full executive, client, and internal-team templates live in report-templates.md. Pull performance metrics/deltas only from performance-analyzer output and financial ratios only from roi-calculator output. If the required computed artifact is absent, return a partial shell plus NEEDS_INPUT and the exact handoff; do not recompute from raw numbers. Pair every metric with compatible source-dated context or mark the comparison pending.

3. Apply visualization and writing guidance — choose the right chart per data point and per audience, and follow the lead-with-outcomes narrative arc. See the visualization recommendations and writing best practices in report-templates.md.

4. Close with recommendations and action items — end every report with concrete next steps; add an opaque owner_ref/deadline action-items table for team and board audiences.

5. Represent the measured cycle without re-deciding it — only preserve an existing Campaign Retro Card from performance-analyzer when it explicitly binds the same campaign_id, opaque creator_ref, current non-forked measured/closed tracker-state ref, locked measurement contract, and precommitted decision-rule ref. Never derive or change the decision in this reporting skill. If the card is absent, mismatched, or stale, emit a bound decision: unknown placeholder plus NEEDS_INPUT for the performance-analyzer artifact. Keep the card outside a client/board-facing copy unless the user requests it.

6. Return, then offer persistence or distribution — return the ref-only report and any Retro Card inline. Offer memory/influencer/report-generator/YYYY-MM-DD-<topic>.md (or memory/ad/report-generator/ for paid runs) for exact WARM-save authorization; the saved artifact keeps client_ref, preparer_ref, contact_ref, and owner_ref unresolved. If the user requests a human-readable audience copy, resolve only the approved refs in-memory for that one explicit audience and discard the mapping after rendering. Before any external send/share/export, require a new exact authorization naming the exact report artifact/version, recipient audience, delivery channel, and allowed identity/asset refs; never infer distribution authority from report generation, transient rendering, or WARM save. Request a separate authorization before promoting any eligible durable fact to memory/hot-cache.md. After an authorized save, offer a handoff to campaign-planner to append the saved artifact reference to the relevant tracker row's evidence_refs; that tracker edit needs a fresh path-and-operation-scoped exact WARM authorization, and this skill neither edits the tracker nor advances its stage.

7. Offer, do not launch, the next-cycle assessment — if the user wants to select the next roster, offer an explicit handoff to fit-scorer with the Retro Card's cited evidence and next_campaign_hypothesis. Do not invoke it automatically and do not translate, invent, or carry forward a STAR/SQS verdict from the qualitative Retro decision.

Example

User: "Create an executive report for Holiday Campaign 2024. roi-ref-01 already reports $50K spend, $165K attributed revenue, 230% arithmetic ROI and 3.3:1 ROAS on its declared verified basis. performance-ref-01 reports 3.5M reach across 15 opaque creator refs and calculated target deltas. The preregistered targets were $100K attributed revenue, 2:1 ROAS, and 2M reach. No segmented breakdown or reallocation rule is included."

Output (excerpt — full template in report-templates.md):

# Holiday Campaign 2024: Executive Summary

## Bottom Line: Campaign Exceeded the Supplied Targets ✅

**ROI: 230% (`roi-ref-01`)** | **ROAS: 3.3:1 (`roi-ref-01`)** | **Attributed Revenue: $165,000**

Metric status: financial values and their calculation/provenance come from `roi-ref-01`; reach and target deltas come from `performance-ref-01`. This report formats those outputs and performs no return or attribution recomputation. No incremental-revenue claim is made.

| Metric | Target | Actual | Status |
|--------|--------|--------|--------|
| Attributed revenue | $100K | $165K | ✅ +65% |
| ROAS | 2:1 | 3.3:1 | ✅ +65% |
| Reach | 2M | 3.5M | ✅ +75% |

### Recommendation

The supplied overall figures clear the supplied targets. A platform, creator-tier, content-format, or budget-change recommendation is `NEEDS_INPUT` because no segmented results or decision rule was supplied; request that evidence before proposing a Q1 reallocation.

Reference Materials

• report-templates.md — full executive/client/team templates, visualization recs, writing best practices, worked example
• skill-contract.md — shared contract and handoff format
• state-model.md — memory tiers and save-path convention
• CONNECTORS.md — free/keyless data recipes per connector category
• performance-analyzer — generates the metrics this report consumes
• roi-calculator — supplies ROI/ROAS figures
• campaign-planner — original plan to compare results against
• content-amplifier — amplification results to report on
• content-quality-auditor — quality gate for the report itself

Next Best Skill

Primary: content-quality-auditor — run the finished report through the publish-readiness gate before it goes to a stakeholder.

Alternates (same report phase / influencer family):

• performance-analyzer — if the report exposes data gaps, re-analyze before re-reporting.
• roi-calculator — recompute return figures if the financial inputs changed.

Termination note (visited-set): if a recommended skill has already been invoked this session, stop and report the chain as complete instead of re-running it. Honor a max chain depth of 3 hops to avoid loops.