oversee: triage task queue — close 16 duplicates and superseded tasks

[fazxes]

Summary

First OVERSEE delegation in the v2 brain era. Addresses task #225 (queue growing) and #226 (brain never uses oversee).

Queue: 72 pending -> 65 pending (9 wontfix-in-active-dir -> done for archiving)

Tasks Closed from Pending (7)

Merged into primary tasks (5)

• fix: stale .next-id silent-drop and autonomy score fabrication bypass #175 -> fix: close costs.json budget-stop poisoning and dirty-clone eval detection (#0125) #174: Both add tests to TestAuthFailureDetection in test_nightshift.py. Same PR scope — the malformed-JSON test (fix: stale .next-id silent-drop and autonomy score fabrication bypass #175) is a natural addition alongside the non-result-event test (fix: close costs.json budget-stop poisoning and dirty-clone eval detection (#0125) #174).
• fix: pentest hardening #0087 -- mktemp guard + pentest report cap + #0169 scope #163 -> overseer: fix ALERT_CONTENT pentest-tag bypass + urgent task for Codex false-green #162: Both are scoring module test additions from the same PR fix: pentest guard gaps + eval scorer rejected-cycle fallback #158 code review. _extract_cycle_fixes empty-fixes test (fix: pentest hardening #0087 -- mktemp guard + pentest report cap + #0169 scope #163) belongs with the mixed accepted+rejected run test (overseer: fix ALERT_CONTENT pentest-tag bypass + urgent task for Codex false-green #162).
• fix: round 6 — all 9 remaining issues patched #124 -> overseer: dedupe auto-release queue #122: Both tasks validate doc snapshot consistency (README, tracker, handoff). Same validate-docs.sh pass, same PR.
• audit: framework 76-session review -- fix staleness and contradictions #196 -> security: close prompt-alert pre-poisoning path (pentest W1) #173: Both add static file entries to PROMPT_GUARD_FILES in lib-agent.sh. One-line additions to the same array.
• fix: configure git identity in TestGitStatusShort CI test #180 -> achieve: fix #0184 -- separate stdout/stderr in pick_session_role #179: Both touch _is_valid_eval_file() in pick-role.py, same PR fix: pentest security hardening -- autonomy first-match, eval validation, unified.md guard #170 origin. CRLF test + stderr warning belong together.

Closed as obsolete (1)

• feat: deepen repo profiler context #78: References "evolve.md Step 8" which does not exist (current evolve/SKILL.md has 6 steps). Also references the multi-agent review panel (PR feat: multi-agent PR review panel (task #0047) #63) which was replaced by unified review (PR feat: unified daemon — agent picks its own role each cycle #107). The concept is valid but the task instructions target non-existent infrastructure.

Closed as low-value (1)

• fix: sessions-since counters parse delegation history from decisions log (#0222) #230: _DELEGATION_ROLE_MAP already covers all 8 current agent types. New agent types require major framework changes where updating a 10-line dict is trivially obvious. Checklist/test overhead exceeds value.

Wontfix -> Done (9 tasks)

Tasks 0077, 0080, 0107, 0111, 0115, 0119, 0127, 0129, 0134 already had status: wontfix with documented rationale. Converted to status: done so the daemon's archive_done_tasks() housekeeping removes them from the active directory. wontfix is not a valid status per the task guide (only pending, in-progress, done, blocked).

Verification

• make check passes (1164 tests)
• No task files deleted
• Every closed task has a documented reason
• Primary tasks updated to absorb merged acceptance criteria

Tasks NOT Closed (kept as genuine work)

Reviewed all 72 pending tasks. The following were examined for possible closure and kept as genuine work:

• feat: production-readiness checker module for Loop 2 #69 (auto-changelog) — not started, valid feature
• fix: target step 0 evaluations at cloned repos #94 (E2E eval wired to daemon) — not done, run_evaluation not called in daemon.sh
• Open PR from previous session requires full review before merge #103 (CI/CD umbrella) — blocked by design, github-issue source, kept
• fix: gate autonomous queue on eval score #104 (CI structured report) — not done, ci.yml has no artifact upload
• docs: close stale eval startup task #105 (sync defaults) — not done, sync omits target/vision_section defaults
• fix: add Phractal eval verification metadata #106 (backlog budget) — not done, SKILL.md has no 0-2 task cap
• fix: round 2 audit — 9 issues across unified daemon #113 (surface healer failures) — valid, || true silently swallows errors
• feat: Python role selector — works for both Claude and Codex #114 (centralize healer default) — valid, shell and Python have independent hardcoded 50
• overseer: decompose stuck integration tasks #120 (list-tasks.sh permissions) — valid, file is not executable
• overseer: dedupe auto-release queue #122 (README consistency checks) — valid, no validation exists
• feat: overseer rewrite — ticket closer that reduces the queue #123 (path-casing helpers) — condition is met (canonical_repo_relative_path used in both worktree.py and owl/cycle.py)
• And all other pending tasks — examined, kept as genuine pending work