[muqsitnawaz]

What

linear tasks --json now gives each issue's assignee their Linear profile picture, as assignee.avatarUrl. The AGI extension's Issues tab draws the assignee from it: the owner asked for real profile pictures instead of colored initials. Before this, the only way to get them was a second linear users read.

The change is one field in ISSUE_FIELDS (assignee { name avatarUrl }). tasks --json already prints the nodes as read.

Run

Rush project (63 issues), from this branch:

$ LINEAR_PROFILE=default ./linear tasks --project 8fa120d1-… --json | python3 -c '…'
63 issues; 62 of 62 assignees have avatarUrl; e.g. {'name': 'Bisma', 'avatarUrl': 'https://public.linear.app/0a580bfd-…/b971b1e4-…'}

Tests

python3 -m unittest: Ran 285 tests, OK (skipped=4). The new test is IssueFieldsTest.test_task_lists_carry_the_assignees_profile_picture.

🤖 Generated with Claude Code

[muqsitnawaz]

VERDICT: APPROVE

Non-author review of the diff (ISSUE_FIELDS, CHANGELOG, test_linear.py).

• Query safety: ISSUE_FIELDS is interpolated at only two sites, paginate_issues (linear:1884, pages of _PAGE_SIZE = 100, linear:1867) and semantic_rank (linear:2269, first=25). avatarUrl is a scalar on User, already queried elsewhere (linear:887, 3132, 4844), so no schema risk. It sits on an assignee object that was already selected, so the added complexity is one scalar per node, nowhere near the 10,000 cap noted at linear:4847. Overview and members queries are separate and untouched.
• Non-JSON output: the human renderers read only assignee.name (e.g. linear:1764). The JSON prints at linear:5380 and 5583 dump nodes as read, so only the JSON gains the field.
• Cache and equality: the embeddings cache keys on id/updatedAt (linear:1704, vectors table), and embedding_text uses title and description only (linear:2607). The issues table stores node JSON per issue (linear:2630). Nodes cached before this change lack avatarUrl until the issue's updatedAt moves. That cache feeds --similar ranking, not tasks --json output (which refetches), so it is harmless. I found no dedupe or equality on assignee dicts.
• Test: python3 -m unittest gives Ran 285, OK (skipped=4). The new test pins the field string in ISSUE_FIELDS and would fail if it were reverted. It is a source-shape assertion, not a live-API check, but it is proportionate for a one-token change.
• CHANGELOG: an ### Added entry under [Unreleased], matching the Keep a Changelog format of the file.
• Not run: the live linear tasks --json call was blocked by the sandbox permission policy, so the real API response is unverified. Adding a documented scalar field is low risk; if the PR body carries no run, a quoted run of that command would be a nice addition (NICE only).