Skip to content

FIX-1904: an item asked of the chief of staff is coded in the right project, and its goal runs alone - #3205

Merged
jhoffner merged 4 commits into
mainfrom
claude/project-thread-hg4517
Oct 13, 2026
Merged

jhoffner merged 4 commits into
mainfrom
claude/project-thread-hg4517

Conversation

@jhoffner

@jhoffner jhoffner commented Oct 13, 2026 •

Copy link
Copy Markdown
Collaborator

Requested by Jake · project thread

Before: on pnpm shift-manager:dev, an item asked of the chief of staff on FSD or Shift Manager never got coded. Best-fit routing hands the post to the EM. The EM filed the coder's task with no project, so the run was refused (no-project). Once it named the project, it wrote a slug as the task's goal. Claude Code sees only that goal of the ask, so it built the team's standing feature brief (greet() in src/greeting.js) instead of the item. The goal check's leg b also couldn't run alone (GOAL_LEG=b started the script with --assets undefined).

After: the EM names the project the ask names, and keeps every file name and value of the ask in the goal. Both items are committed by Claude Code in worktrees of the checkout, cut from main.

How: two sentences in the EM's WORKER.md. In the goal script, leg b builds the pages it starts on, reads the task from the EM's board under the conversation as well as the conversation's own, and has the no-checkout control read the refusal from the run's recorded answer (a board row says errored, not why).

Verdicts (FIX-1904 goal, signed in with ANTHROPIC_API_KEY)

Run Verdict
GOAL_LEG=a PASS: shared fsd, shift-manager on this repository, none private; "Coder: Claude Code, on the credential from ANTHROPIC_API_KEY."
GOAL_LEG=b, before the EM fixes FAIL: no-project, then the coder built greet()
GOAL_LEG=b, this PR PASS: both tasks completed, one worktree each under fsd-runs, from main and not the checked-out branch, committing the item's file with its stamp
GOAL_CONTROL=no-checkout GOAL_LEG=b, this PR FAIL as required, on b:worktrees-in-checkout (0 and 0); each run refused remote-not-allowed, nothing cloned

V1 (spec D1): probeClaudeAccount on a home logged in to Claude Code with a Console key and no variable reads account (apiKeySource: "/login managed key", 1.2 s); an empty home reads none. A claude.ai subscription (OAuth) sign-in was not available here and stays unrun.

shift-manager unit tests: 435 passed. tsc -p goals clean. No changeset: only a Lab's worker file and a goal change.

🤖 Generated with Claude Code

https://claude.ai/code/session_015n8Fc76g745X9Mync4YWcs

GOAL_LEG=b started the script with --assets undefined, so leg b and its
no-checkout control could never run on their own.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_015n8Fc76g745X9Mync4YWcs
@vercel

vercel Bot commented Oct 13, 2026 •

Copy link
Copy Markdown

The latest updates on your projects. Learn more about Vercel for GitHub.

3 Skipped Deployments
Project Deployment Actions Updated
flow-state-dev-docs Ignored Ignored Oct 13, 2026 12:48am UTC
flow-state-dev-kitchen-sink Ignored Ignored Oct 13, 2026 12:48am UTC
flow-state-dev-ui Ignored Ignored Preview Oct 13, 2026 12:48am UTC

Request Review

Copy link
Copy Markdown
Collaborator Author

Second-look / architecture review: approve as-is, nothing blocking.

Verified the premise. startScript passes --assets pages! for the default "start" form (run.mts:90). Leg b calls startScript(env) at line 219, so GOAL_LEG=b alone hit pages === undefined and would spawn with a bad --assets argument. The unconditional build is a real fix, and it is the smallest one.

Is there a better approach? Not one worth the churn. The only alternative is to build lazily inside startScript on first use of the "start" form. That would drop the module-level let pages and the pages! assertion, and the ordering coupling would go away. But it is a bigger diff for a goal script that always runs a leg needing the pages, so it does not earn its scope here. This PR is not a natural moment to refactor.

Cost. One build now runs for a GOAL_LEG=c-only run. Leg c's dev form does not use the pages, but its start form does (run.mts:291). So every selectable leg needs the build and nothing is wasted.

Optional nit (not blocking). The new comment says "Every leg starts the script on the built pages." That is true for a, b and c's start form, but not for c's dev form. Something like "every leg starts the script in start form, which needs the built pages" would be exact. Fine to leave.

Architecture follow-up (no action for this PR). pages! is a non-null assertion on a module-level mutable that is set inside runGoal. That is the underlying hazard, and this PR removes the way it could actually be hit. Not worth a separate issue on its own.


Generated by Claude Code

Asked through the chief of staff, best-fit routing hands the item to the
EM, whose session leads no workstream. It filed the coder's task with no
input.project, so the run was refused (no-project) and nothing was coded.
The EM now names the project the ask names. Leg b reads the task from the
EM's board under the conversation as well as the conversation's own.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_015n8Fc76g745X9Mync4YWcs
The coder reads only the task's goal of what was asked. The EM wrote a
slug ('create-fsd-scratch-file-and-commit'), so Claude Code worked the
team's standing feature brief instead of the item.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_015n8Fc76g745X9Mync4YWcs
…nd the verdicts

A board's row says errored, not why, so refused-by-name could never pass.
Records the signed-in verdicts for legs a and b and the no-checkout control.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_015n8Fc76g745X9Mync4YWcs
@jhoffner jhoffner changed the title Always build shift-manager pages regardless of test legs FIX-1904: an item asked of the chief of staff is coded in the right project, and its goal runs alone Oct 13, 2026
@jhoffner jhoffner self-assigned this Oct 13, 2026
@linear-code

linear-code Bot commented Oct 13, 2026

Copy link
Copy Markdown

FIX-1904

@jhoffner
jhoffner merged commit 5c6d349 into main Oct 13, 2026
19 checks passed
jhoffner added a commit that referenced this pull request Oct 14, 2026
…net 5 says a prompt is a call site

Ledger rows for the epic PR, its amendment, four issue specs and twelve
implementation PRs (three outside the epic but born from it), plus the
wrap's job-surface map and one audit per job that grew. Four fold
candidates filed (FIX-2012..FIX-2015) and one comment on FIX-1986.

Proposed: one sentence in tenet 5. Two of #3205's prompt fixes left
their causes in place and both came back as closure blockers
(FIX-1965, FIX-1971), the same shape as cycle 22's #2768.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MJsSFWmTXycBXJU4s96Pk1
jhoffner added a commit that referenced this pull request Oct 19, 2026
* distill(FIX-1903): cycle 24 ledger for the shift-manager:dev epic; tenet 5 says a prompt is a call site

Ledger rows for the epic PR, its amendment, four issue specs and twelve
implementation PRs (three outside the epic but born from it), plus the
wrap's job-surface map and one audit per job that grew. Four fold
candidates filed (FIX-2012..FIX-2015) and one comment on FIX-1986.

Proposed: one sentence in tenet 5. Two of #3205's prompt fixes left
their causes in place and both came back as closure blockers
(FIX-1965, FIX-1971), the same shape as cycle 22's #2768.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MJsSFWmTXycBXJU4s96Pk1

* distill(FIX-1903): shorter tenet 5 clause; job audit as filed bullets; fix sample count

Tenet 5 keeps only "a prompt is a call site too" and its rule; the
examples move into the cycle entry. The ledger's job-surface and
job-audit tables collapse to a "Filed, not proposed" list (the full
addition list stays in the PR body). The scope count is 18, matching the
table.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MJsSFWmTXycBXJU4s96Pk1

* distill(FIX-1903): drop the cycle 24 lead that retells the (a) table

Folds the subtraction proposed in #3485: the bold result stays, the
owner's full #2768 sentence moves into the cycle 22 cell, and "every run
was refused" / "input became an object" move into the FIX-1965 cell.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MJsSFWmTXycBXJU4s96Pk1

* distill(FIX-1903): scope the tenet line, class every finding, name the approval endpoint

The tenet 5 sentence now applies only when the instruction is right, so a
genuinely missing or wrong prompt can still be fixed in the prompt. Each
cycle 24 row carries its findings' classes, with totals per kind, and the
endpoint note says the owner's merge is the direction approval here.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MJsSFWmTXycBXJU4s96Pk1

---------

Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants