You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
feat: readable shift digest picks + a shadow row that measures the human baseline #579
Shadow shift 1 of the orchestrator pilot (#574) showed the digest can't be reviewed. It listed 6 bare prompt slugs with why: scoped tier notify, which carries no ranking, and a page of all-zero counters. The shadow pilot row also measured the wrong thing: minutes spent reading the digest, rather than the human's own review rate the epic asks for as the baseline.
Plan
Each pick records and shows its prompt title, size, repos, epic/phase and a link, in the digest and in batch watch.
why: adds the planner's actual ranking: queue.md position, or "not in queue.md" with review-minutes, priority and epic phase.
A shadow digest collapses its all-zero counters to one line.
A shadow pilot row lands merges as ? for check-in; the docs say shadow review-minutes means the human's own review time.
Tier: glance — merge mode: auto-merge on green if the Witness passes, post-merge summary (a non-PASS Witness → human /prm).
_batch.survey adds title (the prompt's first # line).
_orchestrator._pick_record(member, brief, proposed=False): the one pick dict (prompt, why, tier, title, size, repos, epic, phase), used by _propose and by dispatch.
_brief.pick_reason: the admission rule, then · and _rank_note(member): queue.md #k, or not in queue.md · ~N review-min · priority P, plus epic E phase X.
_digest: pick_line(task, pick) renders a title-first line with a link to the prompt (on the GitHub Mind main, from the issue: repo) and the details. Brief work and backlog both use it. Older picks without a title fall back to the slug. validate still finds the why:.
_digest.build: a shadow digest prints one - all counters 0 (shadow: nothing dispatched) · review-minutes-planned N line. validate accepts that form only for a shadow digest.
_digest.pilot_row: for shadow, merges → ?. PyAutoMind/batches/AGENTS.md and the batch AGENTS.md: shadow check-in = your own review minutes and merges in the window.
_watch: task lines show the title.
Tests: extend digest/shadow/watch tests; full Brain suite; tenant firewall.
Merged in #580 (4faaa67a) as a glance-tier in-turn merge: Witness PASS, CI green, Heart GREEN.
Digest picks now read by title with a link to the prompt, plus size and repos. why: carries the planner's real ranking. Shadow counters print as one line, and a shadow pilot row leaves merges for the human's own baseline.
Activity
Jammy2211 commentedon Nov 4, 2026
Shipped
Merged in #580 (
4faaa67a) as a glance-tier in-turn merge: Witness PASS, CI green, Heart GREEN.Digest picks now read by title with a link to the prompt, plus size and repos.
why:carries the planner's real ranking. Shadow counters print as one line, and a shadow pilot row leavesmergesfor the human's own baseline.Revert:
git revert -m 1 4faaa67ain PyAutoBrain.🤖 Generated with Claude Code