From macroscope
Run a local Macroscope review with the installed CLI, triage streaming findings, fix the valid ones, and report only the issues you addressed.
npx claudepluginhub joshuarweaver/cascade-code-general-misc-2 --plugin prassoai-macroscope-localThis skill uses the workspace's default tool permissions.
Run a code review using the installed `macroscope` CLI, then triage streaming issues as they arrive. This is the local path for the top-level Macroscope router (`/macroscope` in Claude Code, `/macroscope:macroscope` in Codex) whenever Macroscope correctness review has not already run for the current local `HEAD`.
Creates isolated Git worktrees for feature branches with prioritized directory selection, gitignore safety checks, auto project setup for Node/Python/Rust/Go, and baseline verification.
Executes implementation plans in current session by dispatching fresh subagents per independent task, with two-stage reviews: spec compliance then code quality.
Dispatches parallel agents to independently tackle 2+ tasks like separate test failures or subsystems without shared state or dependencies.
Run a code review using the installed macroscope CLI, then triage streaming issues as they arrive. This is the local path for the top-level Macroscope router (/macroscope in Claude Code, /macroscope:macroscope in Codex) whenever Macroscope correctness review has not already run for the current local HEAD.
This workflow is closed-loop:
Do not stop after triage to ask for confirmation. Do not commit or push.
Auto-detect whether you're on a feature branch so the review covers all changes (committed + uncommitted):
git merge-base --is-ancestor HEAD origin/staging && echo "ON_STAGING" || echo "FEATURE_BRANCH"
ON_STAGING: skip --base and review uncommitted changes only.FEATURE_BRANCH: use --base staging.The codereview command is blocking. It maintains an agent loop for tool calls and a watch stream for progress until the review completes. Run it in the background so you can poll from the foreground:
macroscope codereview --base <base_branch> >/tmp/review-stdout.log 2>/tmp/review-stderr.log &
Keep this background process alive while you poll. It is servicing the remote review agent's tool calls.
Wait for the build/upload/start phase to complete, then extract the review_id from stderr:
grep -m1 'review_id=' /tmp/review-stderr.log
Save this JWT. You need it for polling partial results.
If no review_id appears after a reasonable wait, inspect /tmp/review-stderr.log, surface the failure, and stop.
While the background process runs, poll for incremental results:
macroscope codereview --status '<review_id>' 2>&1
This returns the current status snapshot:
{
"status": "in_progress|completed|failed",
"issues": [...],
"poll_after_seconds": 5,
"is_final": false
}
is_final is false, keep polling and wait poll_after_seconds between polls.is_final is true, the review is done and all issues are final.Maintain a seen-set of issue fingerprints so you only process newly surfaced findings on each poll. Use a stable fingerprint such as:
file:start_line:end_line:category:message
Don't wait for the review to finish. As soon as issues appear in a poll response, start triaging and fixing new findings.
<triage_rules> For each issue, read the affected file and evaluate whether the finding is legitimate. Classify each as:
For each new finding:
Continue the loop:
poll → triage new issues → fix valid ones → verify → poll again
until is_final is true.
Once the review is final:
codereview process exit naturally. If it is still alive after the final snapshot and you no longer need it, stop it cleanly.When you report back:
Do not commit or push.