Hyphenomenon Codex Usage Report

Hyphenomenon Codex Usage Report

Custom skill created by Maggie Lerman.

Produce a period-specific Codex usage report while keeping deterministic evidence, interpretive analysis, preview, and publication permission separate.

Runtime Contract

• Use GPT-5.6 Sol with high reasoning for every interpretive report run.
• If the model/reasoning configuration cannot be verified, stop before writing the analytical narrative and state the required configuration.
• Treat the Hyphenomenon repository's analyzer, validator, sync script, and report templates as the enforceable source of truth. Do not copy their logic into this skill.
• Generation is not permission to mutate production.

Workflow

1. Orient in Hyphenomenon.

• Work from a clean, current Hyphenomenon checkout.
• Read AGENTS.md, ROADMAP.md, CHANGELOG.md, and DOCS/PROJECTS/active/1102_daily-codex-report-artifact-intake.md.
• Confirm the requested start, end, timezone, report slug, title, publication visibility, and output run directory.
• Use explicit ISO boundaries in America/New_York; never infer a day from the current clock alone.

2. Generate deterministic evidence.

npm run intake:report:codex-usage -- \
  --start "<start-iso>" \
  --end "<end-iso>" \
  --timezone "America/New_York" \
  --slug "<report-slug>" \
  --title "<report-title>" \
  --summary "<baseline-summary>" \
  --project-slug "1102" \
  --output "DOCS/intake/runs/reports/<run>/artifact.json"

• Record the emitted analyzer version and source window.
• Require validation totals to reconcile before interpretation.
• Treat completed mcp_tool_call_end records as app/provider invocation evidence.
• Count a skill only when the operator explicitly requests it or a tool input observes a literal SKILL.md load. Do not count available tool/skill catalogs, and do not infer that a loaded skill was applied.

3. Perform the Sol/high analytical pass.

• Read the complete structured artifact before editing public prose.
• Identify the period's actual story: work blocks, repository distribution, verified delivery, output concentration, Agent Stack, Chronicle coverage, anomalies, and uncertainty.
• Lead with verified outcomes and meaningful patterns. Keep raw tool/event counts as supporting telemetry.
• Rewrite the title, summary, and sections for this period when the generated baseline is generic or misleading.
• Preserve deterministic numeric fields unless the analyzer is corrected and rerun.
• Never describe patch churn, commit-operation count, tool volume, app mentions, or Chronicle lane overlap as human productivity or duration.
• Keep analyzer-wide definitions in Method Notes; the template presents them in the metadata disclosure.

4. Validate and inspect data quality.

npm run intake:validate:report -- --artifact "DOCS/intake/runs/reports/<run>/artifact.json"

• Confirm every validation passes.
• Check that app/provider counts are completed invocations and skill rows carry Requested, Loaded, or both.
• Check repository labels, source links, summaries, and public prose for private paths, transcripts, IDs, secrets, or unsupported claims.
• If Chronicle coverage is partial, say so. Citation occurrences, unique cited minutes, and captured frames are different measures.

5. Preview without production mutation.

• For a public existing report, use only the guarded .env.local exact-target update path after a dry-run proves nonDestructiveUpdate=true.
• For a review-only report, use Hyphenomenon's development-only tracked review-artifact loader; do not sync it to the database.
• Review /reports/<slug> and /evidence?type=report at desktop/mobile widths, light/dark themes, keyboard navigation, and print layout.
• Require no runtime errors, console errors, horizontal overflow, or missing report/index links.

6. Gate publication.

• Commit and push the validated artifact, docs, and evidence on a review branch first.
• Production apply is allowed only when the active Hyphenomenon policy independently proves a clean main, a net-new target, schema-neutral additive behavior, no overwrite, a run-local rollback packet, successful production dry-run, route verification, and drift verification.
• Stop for existing targets, private/review-only visibility, schema changes, destructive behavior, ambiguous source coverage, or any failed validation.
• Never publish Chronicle review artifacts or source imagery through this skill without the separate approved Chronicle promotion plan.

Completion Packet

Report:

• source window and timezone
• analyzer version
• Sol/high analytical story and material uncertainties
• artifact and local preview paths
• validation, test, build, and visual-QA results
• database target actually used, if any
• branch/PR status
• explicit statement of whether production or automation changed