# Radar Review

> Triage a literature-radar digest: judge each candidate against what this project is actually doing, and queue only what earns it.

- Skill: `misaka16384/radar-review` (Agent Skill)
- Install (CLI): `npx skillmds@latest add misaka16384/radar-review`
- Raw SKILL.md: https://api.skillmd.com/api/skills/misaka16384/radar-review/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Research & Search
- Author: Misaka16384 (https://skillmd.com/u/misaka16384)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/misaka16384/radar-review

---


# radar_review

## When to use
`magi radar status` reports a digest waiting, or `magi next` says so.

## Method
1. `magi radar status --json`, then read the digest.
2. Read the project first: `magi stats wiki-summary` and the open
   propositions in `threads/`. Relevance is to what is asked here, not the field.
3. Score each candidate by hand. The digest's score is a **rank, not a
   measurement**, and a project with no vector index has none at all — then the
   order carries nothing and every candidate has to be read.
4. `magi search "<title>"` before keeping one: a paper the project already
   covers is not new information, it is a duplicate.
5. `magi radar triage --id <id> --decision accept|dismiss`, ids from the
   digest and `--id` repeatable when the decision is the same. Never hand-edit
   the frontmatter — the triage ledger is what the next run reads.
6. Accepted papers: `magi ingest url "<id>"` to queue them, then hand off to
   the `ingest` skill.
7. `magi radar triage --done` closes the report and prints how many of how many
   you decided. Until you run it, `sync` and `next` keep reporting it.

## Rules
- **Never** accept a paper because it scored well. The score orders the
  reading; the judgement is yours and it is the whole point of this step.
- **Each candidate is its own question.** Judge it against the project, not
  against your reasoning about the ones before it: twenty in, that reasoning is
  the loudest thing in context and the least relevant, and the bar drifts.
- **Never** leave a digest half-triaged without saying so. `--done` prints the
  count; say the same thing in your own report, and where you stopped.
- A citation-gap report asks a different question: whether this project *owes*
  a citation, not whether the paper is nearby. Most nearby papers are not owed.