Weekly Radar Ingestion

Weekly Radar Ingestion

Purpose

Ingest one Ray-reviewed AI Signal Radar report into ai-signal-radar as an approved signal record for future AI agents. Treat the report as canonical input, not as a draft or news clipping. Preserve Ray's reasoning and use additive, deterministic updates.

If details are missing, leave structured fields empty or ask for the missing date. Never fabricate dates, themes, signals, action items, or importance.

Repository Contract

Before editing, inspect the target repository from its root:

• README.md
• schemas/
• existing radars/, data/, themes/, and indexes/
• available scripts under scripts/
• git status --short

Prefer existing repository scripts and schemas when they exist. If a schema is stricter than the recommended metadata below, conform to the checked-in schema unless the user explicitly asks to update the repository schema too.

Read references/repository-contract.md when the repository structure, schema, or theme file format is unclear.

Workflow

1. Detect Track And Report Date

Select an explicit repository track from the approved report and user intent:

• systems for AI systems engineering and infrastructure;
• creation for AI-native creation, human behavior, and creative forms.

Do not combine tracks. Creation reports also require explicit observation window start and end dates.

Determine one YYYY-MM-DD report date using this order:

1. Explicit user instruction.
2. Metadata block in the report, such as date:.
3. Report title or filename containing an ISO date.
4. A clear week-ending date inside the report body.

If no date is reliable, stop and ask for the date. Do not infer a date from the current day unless the user explicitly says the report is for today.

Set week to an ISO week string such as 2026-W25 when useful. Leave it empty if the repository schema does not support it.

2. Store the Original Report

Write the reviewed Markdown report to:

radars/<track>/YYYY/YYYY-MM-DD.md

Preserve content. Normalize only:

• heading levels and spacing
• Markdown list and code-fence formatting
• links
• frontmatter or metadata block

Do not significantly rewrite the report. Do not summarize sections. Do not remove caveats, reasoning, uncertainty, or Ray-specific judgment.

Use templates/report.md only as a formatting reference; do not force a report into that structure if it would rewrite content.

3. Generate Metadata

Write structured metadata to:

data/<track>/YYYY/YYYY-MM-DD.json

Recommended agent-friendly fields:

{
  "schema_version": "2.0",
  "radar_type": "systems or creation",
  "date": "",
  "week": "",
  "title": "",
  "status": "reviewed",
  "reviewed_by": "Ray",
  "themes": [],
  "importance": 0,
  "top_signals": [],
  "action_items": []
}

When the repository already has schemas/radar.schema.json, validate against that schema and use its field names. In the current ai-signal-radar schema, store signal objects in signals, recurring creation forms in patterns, ideas in ideas, and source provenance in source. Creation signals may preserve domain, creator, evidence level, what was created, what AI changed, human behavior change, and a transferable idea. Do not add unsupported properties unless the schema is updated in the same change.

Use templates/metadata.json as the neutral metadata shape for repositories without a stricter schema.

4. Extract Themes

Extract themes only from explicit report labels, metadata, headings, repeated topic framing, or unambiguous signal content. Use lowercase hyphen slugs.

Known examples:

• coding-agents
• agent-runtime
• llm-infrastructure
• evaluation
• observability
• backend-architecture
• ai-platforms
• ai-security
• mlops

If a report has no explicit themes, set themes to [] and continue. If a report introduces a genuinely new explicit theme, create that theme file rather than forcing it into an existing bucket.

Maintain one file per theme under:

themes/<track>/<theme>.md

Each theme file should keep these sections:

• # <theme>
• ## Summary
• ## Related Reports
• ## Notable Recurring Signals

Use templates/theme.md for new theme files. Prefer additive updates to existing theme files and keep report links stable.

5. Update Indexes

Update machine-readable indexes after metadata and theme files are written. Prefer the repository's existing index names and scripts. In ai-signal-radar, run:

python scripts/update_indexes.py

If no repository script exists, maintain:

• indexes/reports.json
• indexes/themes.json

Indexes should optimize future agent retrieval with dates, titles, themes, Markdown paths, data paths, and compact signal references.

6. Validate Consistency

Verify:

• report file exists
• metadata file exists
• metadata date, status, reviewer, paths, and themes are coherent
• creation metadata contains a valid observation window
• every theme reference points to an existing report
• indexes include the ingested report and reference valid files

Use repository validation first when available, for example:

uv run pytest

Then run the bundled consistency helper:

python .codex/skills/weekly-radar-ingestion/scripts/validate_ingestion.py \
  --repo . --track <track> --date YYYY-MM-DD

Fix validation failures before committing.

7. Commit

Generate a concise commit message after validation, such as:

ingest radar YYYY-MM-DD

or:

update themes from radar YYYY-MM-DD

Commit only the ingestion-related files. Push only when the repository has an ordinary configured remote/upstream and push is allowed by the user's request or repository workflow. Never force-push.

Rules

1. Treat the report as approved.
2. Preserve Ray's reasoning.
3. Do not summarize unless a structured field requires extraction.
4. Do not delete information.
5. Prefer additive changes.
6. Maintain deterministic paths and JSON ordering.
7. Optimize for future AI-agent retrieval.
8. Favor machine-readable metadata over prose.
9. Never fabricate themes, dates, signals, action items, or importance.
10. Leave missing fields empty rather than guessing.

Portability Notes

• Specific to the author's current workflow: this assumes Ray-reviewed systems and creation radar reports and an ai-signal-radar repository with explicit tracks under radars/, data/, and themes/.
• Reusable: approved-report preservation, explicit track selection, deterministic track/date paths, schema-first metadata extraction, additive theme/index updates, and validation-before-commit discipline.
• Adapt before reuse: replace reviewer naming, track taxonomy, observation window requirements, repository layout, schema fields, validation commands, commit policy, and theme taxonomy with the target repository's checked-in contract.

Bundled Resources

• templates/report.md: optional Markdown shape for normalized reports.
• templates/metadata.json: neutral metadata template.
• templates/theme.md: new theme file template.
• templates/reports-index.json: fallback reports index shape.
• templates/themes-index.json: fallback themes index shape.
• examples/: sample reviewed reports for normal ingestion, missing themes, and a new theme.
• scripts/validate_ingestion.py: repository consistency checker.