issue-tick dies ImportError in any worktree whose src/ predates a task_workflow symbol (cwd not pinned to main)

Goal

The /issue-tick contract resolves tick_triage.py from the MAIN checkout but runs it with the session's cwd, which for a per-issue session is the WORKTREE. uv run then resolves the explore_persona_space package from the worktree's src/, which is fork-era by design — the Step 5a spec-freshness sync deliberately does not cover src/. A fresh tick_triage.py importing a symbol added to task_workflow.py after the worktree was cut dies with ImportError, exit 2.

Evidence (observed live, 2026-08-24, issue-2254 worktree)

The documented invocation form failed:

REPO_ROOT=$(dirname "$(git rev-parse --path-format=absolute --git-common-dir)")
uv run python "$REPO_ROOT"/scripts/tick_triage.py 2254
-> tick_triage: FAILED for #2254: cannot import name 'open_async_ask' from
   'explore_persona_space.task_workflow'
   (/home/thomasjiralerspong/explore-persona-space/.claude/worktrees/issue-2254/src/explore_persona_space/task_workflow.py)

Note the resolved path: the WORKTREE copy. Confirmed by symbol count — grep -c 'def open_async_ask' gives 1 in the main checkout and 0 in the worktree.

Pinning cwd to main fixes it:

uv run --directory "$REPO_ROOT" python scripts/tick_triage.py 2254
-> STALE-REDRIVE status=followups_running, marker age 3m -- api-error-after-marker

Why it matters

The tick's contract says a non-zero exit is treated as STALE-REDRIVE, which fails toward coverage, so this is not silent. But the cost is real: every fire in an affected worktree loads the full /issue skill instead of making one cheap Bash call, which is exactly the context burn the 2026-06-12 redesign existed to eliminate (#522: seven full re-drives in four hours exhausted a session). It also masks the true verdict — in the observed case the real verdict was an api-error-after-marker STALE-REDRIVE, which happens to coincide, but a HEALTHY or GATE-TRANSITION verdict would have been converted into a spurious full re-drive, and a GATE-TRANSITION conversion would SKIP the phone push and the cron teardown.

The blast radius is every per-issue worktree older than the newest task_workflow.py symbol tick_triage.py imports — which grows monotonically as main advances.

Fix sketch

Change the documented invocation in .claude/skills/issue-tick/SKILL.md (§ Contract — guarded no-op tick) from uv run python "$REPO_ROOT"/scripts/tick_triage.py <N> to uv run --directory "$REPO_ROOT" python scripts/tick_triage.py <N>, so both the script AND the package resolve from main. Audit sibling call sites for the same pattern — any tick-turn or watcher shell-out that runs a main-checkout script from a worktree cwd inherits this. Consider having tick_triage.py assert at startup that the imported task_workflow module path lives under the main checkout, and fail with a diagnostic naming the cwd-pinning fix rather than a bare ImportError.

Provenance

Found while running the mandated first-action triage call during a /issue-tick 2254 fire, 2026-08-24. The skill's existing worktree guidance covers script RESOLUTION (and warns against --show-superproject-working-tree) but not package resolution.