Enter Guide
Entered Runs Explain Themselves
The enter shell now gets a generated run guide without taking ownership of the worktree.
0 milestones 3 tasks 10 agents
2 plan 3 execute 3 review 2 fix
139k peak context
110k median execute
38m 45s total 12m 17s per task
1m 45s plan 16m 21s execute 3m 20s review
21 edits 236 commands
codex harness
This run built the orientation file that appears inside a disposable run worktree when a user opens it with the enter command. The point was not a new driver or a new merge path; it was a small safety surface for the human or agent who is suddenly standing inside a worktree that Gantry owns.
How this walkthrough is structured
Feature
What did the run build and ship?
Gantry still renders a run guide, writes it as a git-excluded status file with safe instruction-name aliases, and refreshes it before an entered shell opens.
Build
How did Gantry structure the work?
The run split the guide into pure rendering, conservative file ownership, and final command wiring, with the gate pressure landing on the content boundary and the shell hook.
Render Orientation Guide
3 agents 12m 39s wall time 139k peak context 1 fix
The renderer is still present as `render_enter_guide` in the command module. It describes the disposable worktree, names the run, shows progress and state, points to status and logs commands, lists the run artifacts, and tells the reader not to drive or merge the run from there. Later Gantry rename work changed the product names and the guide now points at the installed manual, but the pure rendering surface remains.
This sprint carried only the Markdown renderer, with the run facts already resolved and no filesystem effects to reason about. That made the unit narrow enough for a fresh agent: choose the guide structure, reuse existing status wording, and keep the prose harness-neutral. The first gate went red, so the pressure was in making the content and tests line up with the repository's real contracts rather than in integration.
Write Guide Files
2 agents 7m 11s wall time 110k peak context
The writer still creates `GANTRY-STATUS.md`, excludes generated paths from local git status, and links free `AGENTS.md` and `CLAUDE.md` names back to that guide. Tests still cover occupied aliases, tracked guide suppression, stale symlink refresh, idempotent excludes, and the normal write. The original Gantry filename was renamed later, but the conservative ownership behavior is still the current implementation.
This sprint took the already-tested guide string and owned the risky worktree rules separately from the enter command. Its brief was about git cleanliness and file ownership: write the guide only when safe, create aliases only when safe, repair stale generated aliases, and report failures as data. The cut held cleanly at the gate, which fits a unit whose correctness could be covered with focused filesystem cases before any shell path depended on it.
Hook And Document
3 agents 12m 17s wall time 90k peak context 1 fix
The enter path still prepares the guide before launching the user's shell and ignores guide-write failures except for a diagnostic. The command documentation still says enter drops a git-excluded status guide and optional standard-name symlinks when names are free. Some original README and TUI edits have since moved or been extended, but the documented command behavior and the exit-status regression coverage remain in the tree.
The final sprint was deliberately small: call the proven renderer and writer from enter after run resolution, preserve shell behavior, and document the visible result. Its gate went red and then through repair, placing the stress at the integration point where best-effort guide generation could not be allowed to change the user's shell outcome. The review finished without a replan, so the original decomposition was enough once that boundary was fixed.