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
Long-lived and resumed sessions act on stale premises: main has moved, PRs referenced in a
handoff have merged or changed shape, skills have been renamed, plugins have shipped new
versions, and locked plans reference in-flight work that has since landed. Today each session
re-derives freshness ad hoc (manual rebase checks, gh pr view sweeps, grep for renamed
surfaces); nothing owns the "re-anchor this session's assumptions against live reality" pass.
Ledger decision B16 (babysit-prs migration, docs/topics/babysit-prs-migration/PLAN.md
Brief AC7) sequences this skill last in that migration: issue first with the boundary
definition, implementation after user sign-off.
Proposed boundary (draft — for discussion)
The skill re-anchors a SESSION's working assumptions. It does NOT own:
source-control:worktree status — per-worktree/branch staleness inventory (behind-base
counts, PR linkage, dirty state) stays there; the freshness skill may CITE it as an input.
source-control:babysit-prs rescan — per-PR fresh-evidence discipline (GitHub is the
state store; every finding re-verified against the live API) stays in the babysit loop; the
freshness skill never classifies PR feedback.
Candidate in-scope checks: handoff/PLAN premise verification (referenced PRs/issues/branches
still in the stated state), base-branch drift vs the working branch, renamed/retired skill and
plugin surfaces cited by the session's inputs, plugin-version drift between installed and
repo-source versions, memory-tier files whose subjects have since merged.
USER-RESERVED (do not implement until signed off)
Name (verb grammar per docs/conventions naming contract)
Problem
Long-lived and resumed sessions act on stale premises: main has moved, PRs referenced in a
handoff have merged or changed shape, skills have been renamed, plugins have shipped new
versions, and locked plans reference in-flight work that has since landed. Today each session
re-derives freshness ad hoc (manual rebase checks,
gh pr viewsweeps, grep for renamedsurfaces); nothing owns the "re-anchor this session's assumptions against live reality" pass.
Ledger decision B16 (babysit-prs migration,
docs/topics/babysit-prs-migration/PLAN.mdBrief AC7) sequences this skill last in that migration: issue first with the boundary
definition, implementation after user sign-off.
Proposed boundary (draft — for discussion)
The skill re-anchors a SESSION's working assumptions. It does NOT own:
source-control:worktreestatus — per-worktree/branch staleness inventory (behind-basecounts, PR linkage, dirty state) stays there; the freshness skill may CITE it as an input.
source-control:babysit-prsrescan — per-PR fresh-evidence discipline (GitHub is thestate store; every finding re-verified against the live API) stays in the babysit loop; the
freshness skill never classifies PR feedback.
Candidate in-scope checks: handoff/PLAN premise verification (referenced PRs/issues/branches
still in the stated state), base-branch drift vs the working branch, renamed/retired skill and
plugin surfaces cited by the session's inputs, plugin-version drift between installed and
repo-source versions, memory-tier files whose subjects have since merged.
USER-RESERVED (do not implement until signed off)
docs/conventionsnaming contract)re-anchorplugin now exists (0.2.0, PR feat(re-anchor): add script-the-deterministic-work corrector (0.3.0) #293) and is the naturalplacement candidate; alternatives:
session-flow(adjacent to handoff/resume) or astandalone plugin.
Acceptance shape (post-sign-off)
Boundary doc + skill with routing evals distinguishing it from
worktree statusandbabysit-prs; no overlap copy — cross-cites only.No linked issue