Layered run UI, one-command start, and live-testing fixes, accept DL-011 (+DL-014) - #39
Closed
jimutt wants to merge 6 commits into
Closed
Layered run UI, one-command start, and live-testing fixes, accept DL-011 (+DL-014)#39jimutt wants to merge 6 commits into
jimutt wants to merge 6 commits into
Conversation
Four surfaces matched to what each costs: a footer status line, a five-line widget above the editor whose height is invariant across run sizes (unit-tested), transcript cards as entry renderers (TUI-only, outside LLM context), and a board overlay on demand. Surfaces repaint on session_start, turn_end, agent_end, and after every command; they clear when no run is active. start now takes decisions directly and tolerates natural syntax: /dld-goal start DL-014..DL-022 expands the range, derives a slug (dl-14-22), guards preconditions, creates the run, adds one item per decision, and dispatches the first continuation. Positional, flag, and range forms all work. Starting without decisions refuses instead of creating the empty run the user hit live. Co-Authored-By: Claude Opus 5 <noreply@pi.dev> Generated-By: pi 0.84.2
list-taken-ids scanned open PRs for decision IDs and treated every hit as taken. A branch with an open PR containing its own decisions — the normal case after pushing work — collided with itself, and guard-preconditions refused to start a run on decisions the branch owns. A PR whose headRefName matches the current branch is now excluded from the taken set. Verified against the repo that reported it: the same range that produced nine false collisions now reports none. Co-Authored-By: Claude Opus 5 <noreply@pi.dev> Generated-By: pi 0.84.2
From the holistic review and live testing: - parseStartArgs no longer drops the first positional decision - projectRoot resolves the git root so subdirectory sessions see the same .dld/ the scripts write to - onTurnEnd gates on suspension and only re-verifies when new evidence arrived, not on every turn while an item sits in verifying - Dispatch claims the item (implementing) before dispatching so the loop is single-threaded; no more re-dispatch livelock - attempts counts completed attempts without a double bump, so the first failure retries instead of blocking immediately (DL-004) - refreshSurfaces uses the same project root resolution User-driven: - Esc suspends the loop instead of watching it restart: an aborted turn (stopReason: aborted) clears the pending dispatch and suspends until /dld-goal resume. DL-014 amends DL-008. - /dld-goal pause calls ctx.abort() so pausing mid-turn stops the current work, not just the next dispatch. - Bounds and the status line measure active time from the event log's pause/resume markers, not wall-clock since creation. A run resumed after a week no longer shows 703m of nothing. Co-Authored-By: Claude Opus 5 <noreply@pi.dev> Generated-By: pi 0.84.2
Moderates from the review: - Two turn_end handlers merged into one; refreshSurfaces ran twice per turn - /dld-goal resume runs guard-preconditions.sh resume before flipping status, per DL-004 - The dead token first-check dropped; the post-await check is the one that does work - reviewNagged clears on invalidate so a new run's items get their own warning - /dld-goal status renders the board instead of dumping raw state.json - The board component's unused disposed variable removed Simplifications: - stateMutations removed from run-state.ts: no production callers, and the only tested delegation layer was the dead one - cardLines/CardData removed: the loop builds card lines inline, and the unused renderer was the worst of both shapes - The fake's unused surface dropped: registerMessageRenderer, registerTool, sendUserMessage, and their recording maps were only exercised by the fake's own self-test docs/plan/goal-loop.md build order updated to reflect Stages 1 and 2 complete. Co-Authored-By: Claude Opus 5 <noreply@pi.dev> Generated-By: pi 0.84.2
…R stack Co-Authored-By: Claude Opus 5 <noreply@pi.dev> Generated-By: pi 0.84.2
The sandbox proxy blocked 0.84.x, so package.json said ~0.84.2 while package-lock.json had 0.83.0. npm ci requires them to agree. Pinning to ~0.83.0 keeps the lockfile as the source of truth; the peer range is "*" so the extension runs against whatever pi version is installed. Co-Authored-By: Claude Opus 5 <noreply@pi.dev> Generated-By: pi 0.84.2
Owner
Author
|
Superseded by feat/pi-harness-extension — all commits from this branch are merged there. See #41 for the findings log on top. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What
Layered run UI: footer status line, a five-line widget above the editor whose height is invariant across run sizes (unit-tested), transcript cards as entry renderers (TUI-only, outside LLM context), and a board overlay on demand. Surfaces repaint on session_start, turn_end, agent_end, and after every command; they clear when no run is active.
One-command run start: /dld-goal start DL-014..DL-022 expands the range, derives a slug, guards preconditions, creates the run, adds items, and dispatches the first continuation. Positional, flag, and range forms all work. Starting without decisions refuses instead of creating an empty run.
Fixes from live testing and the holistic review: parseStartArgs no longer drops the first positional decision, subdirectory sessions resolve the git root, verification re-runs only on new evidence, dispatch claims the item before sending it, the first failure retries instead of blocking, suspension covers the write path, pause aborts the current turn, and bounds measure active time from the event log rather than wall-clock since creation.
Also: the collision detector no longer flags a branch's own open PR as a collision.
Stacks on #38.