chore(dashboard): remove confirmed dead code in the dashboard core

[iamwhatever]

Removes three symbols (13 lines) from the dashboard core that no code reaches. No behaviour change, no refactor, no reformatting.

Scope: top-level src/kiro_crew/dashboard/*.py only. handlers/, routes/, handlers_instances.py and handlers_system.py are excluded — they are covered separately. Those excluded paths were still scanned as callers, so a symbol consumed by a handler is correctly treated as alive.

Deleted

get_active_dump_file() — crash_dump_store.py:697-699

Exactly one repo-wide occurrence: its own definition. Searched *.py *.ts *.tsx *.js *.md *.json *.yml *.yaml *.toml *.sh *.ps1 including test/, docs/, docs/system-specs/, src/kiro_crew/builtin_skills/, .github/ and pyproject.toml entry points, plus the bare-string form and every getattr / globals() / importlib / dispatch-table / unittest.mock.patch-target path. Not in any __all__. Last touched 44 days ago.

Its docstring ("for passing to faulthandler") invites the reading that this is a half-wired link, so that was checked specifically: faulthandler is already wired through open_dump_file()'s return value — server.py:3525 _dump_file = await asyncio.to_thread(open_dump_file) → server.py:3538 LoopStallWatchdog(dump_file=_dump_file, …). The accessor was never on that path.

The _active_dump_file module global at crash_dump_store.py:60 is deliberately kept, even though it now has no production reader. open_dump_file publishes it at :518 to hold the DumpFile — and therefore faulthandler's fd — alive, and test/test_crash_dump_store.py reads and restores it directly (:59, :72, :603, :617, :620, :624, :644) including an assertion that open_dump_file must publish it. Removing it would reintroduce an fd-lifetime bug.

DashboardState.mark_notifications_read() — state.py:6626-6628

Exactly one repo-wide occurrence: its own definition.

Its docstring says "called when client opens notification panel" and nothing calls it, so the lifecycle pair was traced end to end before deleting. _unread_count is incremented at state.py:6591 and reset at state.py:6739 by the live clear_notifications(), so this was not the only reset path. The actual read flow is per-note acked — ack_notification, POST /api/notifications/ack-all (handlers/messaging.py:883), unack — and the frontend's "mark all as read" dispatches ackAllNotifications() → /api/notifications/ack-all, which flips acked; it never asks for a server-side unread reset. The badge is derived client-side (website/src/App.tsx:733, NotificationFeed.tsx:217). The server-side unread field (handlers/messaging.py:814) is consumed only by test mocks.

_read_theme_text() — theme_validate.py:1605-1606

Exactly one repo-wide occurrence: its own definition. No string or getattr dispatch; theme validation does not build a by-name reader table. Not in __all__, docs/, docs/system-specs/, or the theme-pack-authoring skill. Last touched 37 days ago.

Because theme_validate.py is a security surface (path traversal, symlink refusal, CSS denylist), this was checked as a possibly-bypassed validation step rather than assumed dead. It is not one: every live asset / overlay / topbar route reads through the hardened _read_theme_bytes_nolink, which re-checks the symlink (TOCTOU-safe) and then decodes with errors="replace" — consistently, everywhere. _read_theme_text was a plain read_text that skipped the symlink recheck. Nothing routes through it and nothing should, so this removes an unused weaker reader.

Reported, not deleted

_REASONING_EFFORT_VALUES (chat_persistence.py:154) — dead, but withheld because the fix spans two scopes. It has zero references of any kind (no import, no dotted access, no patch target string) and is a duplicate binding of the same EFFORT_VALUES object already bound live nine lines above as _REASONING_EFFORT_FALLBACK (:145); the "back-compat caller" its comment invokes does not exist anywhere in the repo. But src/kiro_crew/effort.py:42 describes EFFORT_VALUES as the "Single source for _REASONING_EFFORT_VALUES", so deleting the alias alone would leave that comment naming a symbol that no longer exists. effort.py is outside this PR's file scope, so the alias and that one-line comment fix should land together in a follow-up rather than be split.

PIN_SCOPES (tailnet.py:845) — not dead code, a drift hazard. It has zero references, but the validator it should back exists as an inline duplicate literal ("node", "login") at config/loader.py:3036. Two independent definitions of the allowed pin-scope set on a security path is the finding; the fix is to point the validator at PIN_SCOPES, not to delete it.

Deferred

Symbol	Reason
_mutate_tags_locked (chat_tags.py:73)	Unreachable with zero cascade, but its sole commit landed 2026-08-08 (22 d) — inside the 30-day gate, and it reads as a new abstraction whose handler migration may still be in flight.
broadcast_browser_event (state.py:9268)	Statically dead (sole caller removed 2026-08-13) with no frontend consumer — browser_event appears nowhere in website/src. Withheld only because unmerged branch refactor/dashboard-state-boundaries relocated this exact method into websocket_hub.py on 2026-08-29; deleting it here would collide.
_PROJECT_LOCAL_SEGMENTS (terminal_commands.py:365)	Unused, and the shared-definition guarantee its comment claims is actually delivered by _is_project_local at :367, so no security coupling is lost. But last touched 3 d ago by an in-flight security fix.
_bump_revocation_gen (token_auth.py:94)	Inert alias whose comment is provably stale (zero importers, zero patch targets); the public bump_revocation_gen is fully live on the logout path. Last touched 20 d ago — gate clears 2026-09-08.
17 test-only symbols	Referenced only by test fixtures that use them as state seeding (save_folders, save_tag_boards, ws_client_count, enable_yolo, check_token_ip, reset_cache, extract_claims_from_token, build_transfer_bundle, and 9 more). Deleting them means rewriting fixtures that reach unrelated suites — refactor, not deletion.

How the candidates were found

Three independent scanners, then a confirmation protocol on the intersection:

Pass	Mechanism	Raw findings in scope
vulture --min-confidence 60	heuristic	342 (238 func/method/class)
AST	defs in scope minus every Name(Load) / Attribute / import target across all repo .py files	9
tokenize	real NAME tokens repo-wide with docstrings/comments/strings excluded; absent from every file but its own	650

The three-way intersection was 9 symbols; the protocol (repo-wide multi-filetype search, string-dispatch sweep, frontend-consumer check, public-surface check, 30-day gate, test-only rule, half-wired-lifecycle rule, security-surface rule) reduced that to the 3 deleted here.

A fourth pass split all 238 vulture func/method/class findings by production-vs-test references: 24 had zero production references outside their own file, of which 17 were test-only and 2 were module dunders (__getattr__ / __dir__, called by the import protocol).

Vulture over-reports roughly 25x on this surface because nearly every api_* symbol it flags is an aiohttp route handler referenced from handlers/ or routes/, and the constants it flags are imported by handlers/*. Four of its 90%-confidence "unused import" findings are also false — NudgeLoop, LLMEvent, CronJob and typing are each used inside a quoted annotation it cannot see (chat_handlers.py:3338, chat_utils.py:363, cron_inject.py:70, loop_watchdog.py:72).

Verification

• Affected-scope pytest: 1119 passed, 0 failed — crash-dump store, theme install / CSS security / config / coverage / authoring-skill, notification settings / bus / phase5 / legacy-actions / push, dashboard, handlers-messaging coverage, dashboard-state-ws, effort.
• flake8, isort --check-only and mypy clean on all touched files.
• black --check flags two reformats (crash_dump_store.py ~119, theme_validate.py ~291) that reproduce identically on the unmodified tree — local version drift, pre-existing, untouched.
• Two local reviewer lanes ran on the diff before it was pushed. One returned no findings after failing every falsification attempt; the other raised the _REASONING_EFFORT_VALUES coupling, which is why that symbol was withheld from this PR.

[github-actions]

Design Review (Fable 5) — ✅ PASS

Design-level review of f6ad484c7c19e678ccbef0fc67aafac18dc354cb — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

All three symbols have zero remaining references repo-wide, the _active_dump_file global is correctly kept (still published at crash_dump_store.py:518), and _unread_count retains its live reset path via clear_notifications (state.py:6739). The deleted _read_theme_text was indeed the weaker reader (no symlink recheck), so its removal tightens rather than loosens the theme security surface. The claims in the description match the diff exactly; nothing undocumented is smuggled in.

Design-Verdict: PASS

Three genuinely dead symbols, each deadness claim independently verified against the tree; removing the weaker unhardened theme reader is a strict improvement.

[DESIGN-REVIEWED] f6ad484

[github-actions]

First Principles Review (Fable 5) — ✅ PASS

Premise-level review of f6ad484c7c19e678ccbef0fc67aafac18dc354cb — why this exists and whether the shipped surface is the smallest honest version. Updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

All three deletions verified independently at HEAD: zero remaining references to any deleted symbol anywhere in the repo, _active_dump_file is still published at crash_dump_store.py:518 and read by test/test_crash_dump_store.py, the live unread-reset path (state.py:6739 in clear_notifications) survives, and the hardened reader _read_theme_bytes_nolink remains the sole theme-read path (4 uses in handlers/themes.py). The change is pure subtraction with no riders. Final review:

First-Principles-Verdict: PASS

Three dead symbols deleted, each with a counted zero-consumer proof that holds at HEAD; nothing rides along and nothing weakens a live path.

What this change ships

Intent: shrink the dashboard core by deleting three symbols no code reaches — a pure SUBTRACTION (neither fix nor addition).

1. get_active_dump_file() removed from crash_dump_store.py — justified; 0 references remain, faulthandler wiring uses open_dump_file's return value, and _active_dump_file (the fd-lifetime keeper) is correctly kept.
2. DashboardState.mark_notifications_read() removed from state.py — justified; 0 callers, and the live reset path (clear_notifications, state.py:6739) plus the acked-based flow already cover the job. Grepped mark_notifications_read: 0 matches.
3. _read_theme_text() removed from theme_validate.py — justified; 0 callers, and it was a weaker duplicate of the hardened _read_theme_bytes_nolink (15 occurrences, all live paths route through the hardened one), so deleting it removes a bypass-shaped spelling.

Watch

• Two declared siblings stay behind with named reasons: PIN_SCOPES (tailnet.py:845) still has a drifting inline duplicate ("node", "login") at config/loader.py:3036, and _REASONING_EFFORT_VALUES (chat_persistence.py:154) is dead pending the effort.py:42 comment fix. Both deferrals are grounded and scoped; just don't let them expire silently.

[FIRST-PRINCIPLES-REVIEWED] f6ad484

[github-actions]

GPT 5.6 Review — ✅ no blocking findings

GPT 5.6 completed its review of f6ad484c7c19e678ccbef0fc67aafac18dc354cb and found no blocking issues.

This comment is updated in place on each push.

Review details

No findings.
[GPT-REVIEWED] f6ad484

False positive or not applicable? A repository writer can comment:
/ai-review override gpt f6ad484c7c19e678ccbef0fc67aafac18dc354cb: <one-sentence reason>

[github-actions]

Opus 4.8 Review — ✅ no blocking findings

Reviewed f6ad484c7c19e678ccbef0fc67aafac18dc354cb — this comment is updated in place on each push.

Review details

All three removed symbols have zero references anywhere in the repository. The PR cleanly removes confirmed dead code — no callers can crash, and dead-code cleanup is a deterministically-owned category regardless.

No findings.

[OPUS-REVIEWED] f6ad484

Verdict parsed from the review's SHA-scoped output markers for commit f6ad484c7c19e678ccbef0fc67aafac18dc354cb.

False positive or not applicable? A repository writer can comment:
/ai-review override fable f6ad484c7c19e678ccbef0fc67aafac18dc354cb: <one-sentence reason>

[bolichen97]

Independently verified every removed symbol (grep across full repo + PR head, string/dynamic/registry lookup checks) has zero remaining references anywhere -- no dangling handler/registry entry (the #6876 defect class). CI all green. Approving.