feat(session): filter comment list by author

[abgregs]

What

Adds --author <name> and --no-author filters to hunk session comment list, letting the CLI narrow inline review comments by author — mirroring the existing --file and --type filters.

Why

In an agent-assisted review, an agent and a human both leave notes on the same session, so hunk session comment list --json returns a mixed set. Today the only way to isolate one author's notes is to pipe the JSON through jq (jq '[.comments[] | select(.author=="pi")]'), which pushes a conceptually one-flag filter into every consumer. A native --author <name> flag — plus a --no-author companion for untagged notes — keeps --json consumers thin and matches the ergonomics of --file / --type.

How it works

The filter threads through the same layers as --type, with the actual filtering in a single place (the daemon list handler):

• core/cli.ts — --author <name> / --no-author options, parsed and forwarded into the command input
• core/types.ts, session/protocol.ts — author / noAuthor on the command input and wire shapes
• hunk-session/cli.ts — the client includes them in the request
• session-broker/brokerServer.ts — one post-branch filter narrows the result by author, covering both the live-comment and review-note paths (both carry author)

--author <name> matches that author exactly; --no-author returns only comments with no author tag — comments added via comment add/apply without an author. (Human c notes are auto-tagged author: "user", so they're matched by --author user.) Neither flag returns everything. Documented in docs/agent-workflows.md.

Testing

• Parse (core/cli.test.ts): --author pi → { author: "pi" }; --no-author → { noAuthor: true }.
• Filter (brokerServer.helpers.test.ts): for a mixed-author comment set, --author returns only the matching author, --no-author returns only the author-less, and no flag returns all.
• bun run typecheck, bun run lint, and bun run format:check clean.
• Verified end-to-end on macOS (screenshots below).

Screenshots

End-to-end via hunkdev on this branch — drag each PNG into the GitHub editor to upload it in place:

1. Add one --author pi comment and one untagged comment

2. hunk session comment list --author pi → only pi's

3. hunk session comment list --no-author → only the untagged one

4. hunk session comment list (no filter) → both

Fixes #489

[greptile-apps]

PR author is not in the allowed authors list.

[benvinegar]

Don't love --no-author meaning "human notes", but I guess we can go with it. Wish there was better / clearer language ... I feel agents would perform better.

[abgregs]

Don't love --no-author meaning "human notes" ... Wish there was better / clearer language

Yeah agreed the language is a bit confusing, and I just realized my PR description was wrong about who actually lands in --no-author (details below). On the naming itself, I stuck with what #489 proposed, but my original description said --no-author "returns only comments with no author tag (human c notes, or agent comments added without --author)" mistakenly implying human c notes land in --no-author.

For context on --no-author, #489 asked to "treat a missing author field as a separate value" via a dedicated flag ("two flags keeps the matcher unambiguous"). Re-tracing the code, --type filters by writer category (the source field — ai / agent / user), while --author filters the optional identity label within a category. So --no-author means only "no identity label" — nothing about human vs agent. Human c notes are actually auto-tagged author: "user", so --author user matches them; the --no-author set is just comments without an author tag.

I've corrected the PR description accordingly so --no-author now reads "no author tag," which holds regardless of where we land on naming. Let me know if that tracks.

Sticking with the --no-author naming, I could work in your idea to improve the agent-facing language by sharpening it in these places:

• help text — becomes "only comments with no author tag"
• both agent docs — docs/agent-workflows.md and skills/hunk-review/SKILL.md get the same one-sentence model: --type = writer category (source), --author = identity tag within it; human c notes are tagged user, so use --author user for those
• the skill's command synopsis — noticed SKILL.md still lists comment list with only --file/--type, so an agent following the recommended workflow never learns the new flags exist; I'd add them there

Happy to give that a shot if it sounds right to you.