docs(pr-flow): make the In Progress / In Review card moves runnable, with a check

[cliffhall]

Closes #2470

Agents following /pr-flow kept leaving the card in Todo. Step 1 had a command for the assignment only, and step 6 ended on a trailing sentence with no command. Seen on #2407.

Changes (.claude/skills/pr-flow/SKILL.md only)

• Step 1 now opens with its own bold rule: Step 1 is two actions — assign the issue, and move its card to In Progress. Both happen before you branch. The code block assigns the issue, moves the card and prints the card's Status at the end. The step is done only when that last line reads card: In Progress.
• Step 6 replaces the trailing sentence with the same block for In Review, plus the same rule: the step is done only when the PR is linked and the card reads In Review.
• No /board-ops option ID is copied. The project node ID, the Status field ID and the option ID are all looked up by name when the command runs (gh project view / field-list). The card is found through issue.projectItems, as board-ops does, so an option recreated after a deletion still resolves.
• Failures are loud. A missing ID means nothing gets edited and the command prints which lookup came back empty. A final status that doesn't match prints this step is NOT done.

Verification

• I ran the block live on this issue, in bash and zsh. It moved the card and printed card: In Progress. I also mistyped the column name on purpose: it printed lookup failed (… option='' …) — nothing edited and then card is 'In Progress', not 'In Progres' — this step is NOT done.
• Step 6's block runs right after this PR opens, to move pr-flow: make the In Progress / In Review card moves part of each step's commands, with a check #2470 to In Review.
• npm run verify:skills passes and npm run local:gate is green.

Eval case (item 5): not added, because no case can observe the move

scripts/skill-eval.mjs records only Skill tool calls and limits the model to Read, Glob, Grep and Skill (ALLOWED_TOOLS). No first-move or chain case can see a gh project item-edit. A chain into board-ops would measure the wrong thing, too: step 1 no longer sends the model there. The description didn't change, so skills:eval wasn't re-run.

🤖 Generated with Claude Code

[Copilot]

Copilot review overview

🟡 Changes recommended

Assignment failures can be masked, and the Step 6 command does not support v1’s board.

Get a fresh assessment by requesting another Copilot review.

Review effort: Balanced
Findings: 2

Open (2)

• Completion check omits required assignee verification · New
• Step 6 hardcodes board #28 for v1 issues · New

What changed in this PR

Updates the PR workflow so board status transitions are executable and verifiable.

Changes:

• Moves cards to In Progress during setup and In Review after PR creation.
• Resolves board IDs dynamically and verifies resulting status.

File	Description
.claude/​skills/​pr-flow/​SKILL.md	Adds runnable board-transition commands and checks.

[cliffhall]

Copilot round 1: 2 findings, both fixed in cfdbe57 (replies are in each thread).

• Completion check omits assignee verification: step 1's success line now also requires the assignment to have succeeded.
• Step 6 hardcodes board Add tab and approval flow for server -> client sampling #28: both blocks now take a BOARD variable (11 for a v1 issue).

Re-ran local:gate: green. Requested round 2.

[Copilot]

Copilot review overview

🟡 Changes recommended

The fallback guidance incorrectly directs v1 issues toward board #28 instead of the selected board.

Get a fresh assessment by requesting another Copilot review.

Review effort: Balanced
Findings: 1

Open (1)

• Incorrect v1 board fallback maps BOARD=11 to #28 · New

Resolved since last review (2)

• Step 6 hardcodes board #28 for v1 issues
• Completion check omits required assignee verification

[cliffhall]

Copilot round 2: 1 finding, fixed in 412944d. The no-card fallback now names the selected board ($BOARD) instead of #28. local:gate is green. Requested round 3.

[Copilot]

Copilot review overview

🟢 Approval recommended

The workflow now satisfies issue #2470 without duplicating unstable board option IDs.

Review effort: Balanced
Findings: None

Resolved since last review (1)

• Incorrect v1 board fallback maps BOARD=11 to #28

[cliffhall]

Copilot review closed. Round 3 was clean ("Approval recommended", no findings, no inline or suppressed comments), so the loop stopped on the first clean round. In total: 3 rounds, 3 findings, all fixed.