fix: board_create_feature refuses a same-title open duplicate

[mabry1985]

Found live, in dogfooding

Spun up a fresh team on portfolio-plugin via Roxy (portfolio_spinup_team) and dispatched one feature. The team's board ended up with two beads for the same work, twice:

bd-2kx / bd-8or   "Create PROTO.md grounding doc..."     (the team's own onboarding task)
bd-1bf / bd-2h6   "Rollup surfacing: priority + stuck"   (my dispatched task)

The team's own audit log confirms the root cause — its own agent called board_create_feature twice, ~7-9s apart, same title each time:

20:02:13  board_create_feature(title="Create PROTO.md grounding doc...")
20:02:20  board_create_feature(title="Create PROTO.md grounding doc...")   <- same title, 7s later
20:02:51  board_create_feature(title="Rollup surfacing...")
20:03:00  board_create_feature(title="Rollup surfacing...")               <- same title, 9s later

board_create_feature had zero dedup — every call just creates a new bead, no matter how many times it's called with the same title.

Why this isn't portfolio-plugin's existing dedup

portfolio_dispatch already refuses a same-title re-dispatch (portfolio-plugin#25/v0.14.6) — but that guard lives on the PM's dispatch call, one tier up. It can't catch a board's own agent (its onboarding pass, or its own reasoning about a task it just received) calling board_create_feature twice within a single turn — which is exactly what happened here, on both the onboarding task and the dispatched task.

Fix

Add the same class of guard, one tier down, at board_create_feature itself — so it protects every caller, not just PM dispatches:

• Refuse when a feature with the same (normalized) title is already open (backlog/ready/in_progress/in_review/blocked).
• force=true creates a second copy anyway.
• A store read failure never blocks creation (better a possible dup than a stuck board) — matches portfolio_dispatch's own failure posture.

Tests

tests/test_board_create_feature_dedup.py — pure _norm_title/_open_duplicate unit tests, plus tool-boundary tests (refuse / force-override / recreate-after-done) via a lightweight fake store (no real br CLI needed, matching this repo's existing test style for store-adjacent logic).

Gate: ruff check . && ruff format --check . && pytest -q — 244 passed (237 + 7 new).

Version bumped 0.26.1 → 0.27.0 (new capability — a force param + refusal behavior on an existing tool).

🤖 Generated with Claude Code

[protoquinn]

👀 Quinn is reviewing — verdict (PASS / WARN / FAIL) + findings to follow.

[protoquinn]

QA Audit — PR #63 | fix: board_create_feature refuses a same-title open duplicate

VERDICT: WARN (preliminary — CI is queued; re-evaluating on terminal-green)

---

CI Status

• test: queued ⏳

Diff Review

• Adds dedup guard to board_create_feature: same-normalized-title open feature → refuse with helpful message, force=True bypasses. __init__.py L82-180.
• Three clean helpers: _TERMINAL_STATES, _norm_title, _open_duplicate — scoped, documented, correct.
• Store read failure never blocks creation (except BoardError → existing=[]) — matches stated fail-safe posture.
• Version bump 0.26.1 → 0.27.0 consistent across pyproject.toml and protoagent.plugin.yaml.

Observations

• LOW: clawpatch structural review unavailable for this repo (checkout cache miss) — reviewed from diff directly; diff is self-contained.
• INFO: 7 new tests (tests/test_board_create_feature_dedup.py) cover normalization, dup detection, terminal skip, refusal, force-override, recreate-after-done. No real br CLI dependency.
• INFO: Exact-match-only dedup is the right trade-off — false-positive silently drops real work; PR description makes this explicit and intentional.

— Quinn, QA Engineer

[protoquinn]

Submitted COMMENT review on protoLabsAI/projectBoard-plugin#63.