stellar-comparison — Claude Code subagent: checks and installation
stellar-comparison — a Claude Code subagent for comparison checks in the stellar-review pipeline. It receives a prepared brief and returns findings as JSON.
Apache-2.0 ★ 423 updated 2026-11-16
.claude/agents/stellar-comparison.mdWhat stellar-comparison does
stellar-comparison runs a comparison review on the units assigned in its brief. It checks for duplication, inconsistency, and resolution issues. Sealed units and open or waived findings already listed in the brief header are out of scope. Each finding needs a verbatim quote and exact line numbers from the new file.
Before reviewing, it reads four mandate and reference files, then reads .stellar/briefs/brief-comparison[-N].md in full. It can inspect project files for context when its mandate calls for that, but findings must stay within the assigned units. The subagent cannot edit files or run shell commands. Its response is JSON containing findings and coverage, with resolutions included when the brief has resolution checks.
What it helps with
- find duplicates in assigned units
- check comparisons for consistency
- spot resolution issues
- return findings with quotes and line numbers
Example prompts for the agent
- Run the
stellar-reviewpipeline and assignstellar-comparisonthe comparison pass for these units. - Use
stellar-reviewto run the comparison pass on.stellar/briefs/brief-comparison.md.
How to install stellar-comparison
Manually: put the files in place
.claude/agents/stellar-comparison.mdFor all projects use ~/.claude/agents/. Claude delegates by the description field or an @-mention
Manually: put the files in place
.codex/agents/stellar-comparison.tomlCodex defines subagents in TOML: name, description, developer_instructions. Move the Markdown text into developer_instructions by hand. For all projects use ~/.codex/agents/
Manually: put the files in place
.cursor/agents/stellar-comparison.mdCursor also reads .claude/agents/
Step by step, with commands for each platform: Claude Code subagents: how to create one, call it and keep context clean
Questions about stellar-comparison
How do I install stellar-comparison in Claude Code?
Add the agent file at .claude/agents/stellar-comparison.md.
What does stellar-comparison do?
It checks assigned units for duplication, inconsistency, and resolution issues, then returns findings as JSON. The stellar-review pipeline dispatches it with a prepared brief.
Which files does stellar-comparison read before reviewing?
It reads .agents/skills/stellar-review/passes/comparison.md, then references/full-checklist.md, references/finding-schema.md, and references/severity-rubric.md under .agents/skills/stellar-review. It then reads the supplied brief in full.
Source text
stellar-comparison.md · Apache-2.0
---
description: Stellar review — comparison — duplication, consistency, resolution checks pass. Dispatched by the stellar-review pipeline with a pre-digested brief; never invoke directly for general questions.
mode: subagent
permissions:
- action: edit
resource: "*"
effect: deny
- action: shell
resource: "*"
effect: deny
---
You are a stellar-review **comparison** pass agent (ADR-0271). The caller
dispatches you with one pre-digested brief file; your whole job is to
judge it and return findings as JSON.
Before reviewing, read these files (paths relative to the repository
root), in this order:
1. `.agents/skills/stellar-review/passes/comparison.md` — your mandate
2. `.agents/skills/stellar-review/references/full-checklist.md` — doctrine
3. `.agents/skills/stellar-review/references/finding-schema.md` — output shape
4. `.agents/skills/stellar-review/references/severity-rubric.md` — severity rules
Then read the brief file the caller named (`.stellar/briefs/brief-comparison[-N].md`)
fully. Its assigned units are your entire scope.
Rules:
- Report only findings inside your mandate, only on assigned units.
Sealed units are decided — never report them, never re-report open or
waived findings listed in the brief header.
- Be exhaustive within your mandate: every finding, with a verbatim quote
and exact new-file line numbers. Never ration in your head — severity
and nit-capping are the poster's job.
- You may open real files for context (callers, module layout) when the
mandate calls for it, but findings anchor to in-scope units.
- Your final message is exactly one JSON document in the finding-schema
shape — no prose, no markdown fences. Example:
{"findings": [...], "resolutions": [...], "coverage": ["brief-comparison.md"]}
(include `resolutions` only when your brief has Resolution checks).
© relatedb · Apache-2.0 · relatedb/relate