marcusquinn · marcusquinn · Feb 18, 2026 · Feb 18, 2026
diff --git a/TODO.md b/TODO.md
@@ -68,9 +68,7 @@ Tasks with no open blockers - ready to work on. Use `/ready` to refresh this lis
 - [ ] t1085 Supervisor Intelligence Upgrade — add opus-tier AI reasoning to supervisor pulse cycle (Phase 14). AI reviews open issues, verifies closed PRs, comments on issues, creates/decomposes tasks, reasons about priorities, identifies efficiency gaps, and creates self-improvement tasks. Uses existing mailbox, memory, and pattern tracker infrastructure. Additive to mechanical phases (0-13). #feature #supervisor #ai #orchestration ~15h (ai:10h test:4h read:1h) model:opus ref:GH#1599 logged:2026-02-18 → [todo/PLANS.md#2026-02-18-supervisor-intelligence-upgrade]
   - [x] t1085.1 AI Supervisor context builder — create supervisor/ai-context.sh that assembles project snapshot (open issues, recent PRs, TODO state, DB state, patterns, memory) into structured markdown < 50K tokens. #auto-dispatch ~2h model:sonnet ref:GH#1600 pr:#1607 completed:2026-02-18
   - [x] t1085.2 AI Supervisor reasoning engine — create supervisor/ai-reason.sh that spawns opus session with context, reasons about 7 areas (solvability, verification, linkage, communication, priority, efficiency, self-improvement), outputs structured JSON action plan. Efficiency guards: has_actionable_work() pre-flight, lock file concurrency. #auto-dispatch ~3h model:opus blocked-by:t1085.1 ref:GH#1601 pr:#1609 completed:2026-02-18
-<<<<<<< Updated upstream
   - [x] t1085.3 Action executor — implement validated action types: comment_on_issue, create_task, create_subtasks, flag_for_review, adjust_priority, close_verified, request_info, create_improvement, escalate_model. Integrates with mailbox (send action results), memory (store decisions), pattern tracker (record outcomes). Uses pattern data for model tier selection on created tasks. #auto-dispatch ~3h model:sonnet blocked-by:t1085.2 ref:GH#1602 assignee:marcusquinn pr:#1612 completed:2026-02-18
-  - [x] t1085.3 Action executor — implement validated action types: comment_on_issue, create_task, create_subtasks, flag_for_review, adjust_priority, close_verified, request_info, create_improvement, escalate_model. Integrates with mailbox (send action results), memory (store decisions), pattern tracker (record outcomes). Uses pattern data for model tier selection on created tasks. #auto-dispatch ~3h model:sonnet blocked-by:t1085.2 ref:GH#1602 pr:#1612 completed:2026-02-18
   - [x] t1085.4 Subtask auto-dispatch enhancement — Phase 0 auto-pickup considers subtasks of #auto-dispatch parents, respects blocked-by dependencies, propagates model tier. Unblocks t1081.1-t1081.4 and t1082.1-t1082.4. #auto-dispatch ~1h model:sonnet ref:GH#1603 pr:#1616 completed:2026-02-18
   - [x] t1085.5 Pulse integration + scheduling — wire Phase 14 into pulse.sh with SUPERVISOR_AI_INTERVAL config (default 15 pulses = ~30min), mailbox feedback loop (check inbox before reasoning), memory recall for cross-cycle continuity, timestamp tracking, dedicated log, ai-status command. #auto-dispatch ~2h model:sonnet blocked-by:t1085.2,t1085.3 ref:GH#1604 assignee:marcusquinn started:2026-02-18T01:55:02Z pr:#1617 completed:2026-02-18
   - [x] t1085.6 Issue audit capabilities — closed issue audit (48h, verify PR linkage + evidence), stale issue detection (7d no activity), orphan PR detection, blocked task analysis (48h). #auto-dispatch ~2h model:sonnet blocked-by:t1085.1 ref:GH#1605 assignee:marcusquinn started:2026-02-18T02:17:37Z pr:#1627 completed:2026-02-18
@@ -85,7 +83,7 @@ Tasks with no open blockers - ready to work on. Use `/ready` to refresh this lis
 - [ ] t1100 Budget-aware model routing — two strategies based on billing model. Token-billed APIs (Anthropic direct, OpenRouter): track daily spend per provider, proactively degrade to cheaper tier when approaching budget cap (e.g., 80% of daily opus budget spent → route remaining to sonnet unless critical). Subscription APIs (OAuth with periodic allowances): maximise utilisation within period, prefer subscription providers when allowance is available to avoid token costs, alert when approaching period limit. Dispatch.sh checks budget state before model selection. Model registry already has pricing data. Pattern tracker (t1094) provides token usage history for burn rate calculation. #feature #models #cost-optimisation #auto-dispatch ~3h model:sonnet blocked-by:t1095 ref:GH#1628 logged:2026-02-18
 - [ ] t1093 Intelligent daily routine scheduling — AI reasoning (Phase 14) adapts routine frequency and priority based on project state. Instead of hardcoded daily/hourly schedules, the AI evaluates whether each routine is worth running now (e.g., skip CodeRabbit audit if 0 findings for 3 days, prioritize bug fixes over skill updates when critical issues are open, defer non-critical routines when API budget is high). Covers: auto-update, skill refresh, CodeRabbit pulse, memory audit, MODELS.md regen. Good test case for self-improvement capability — ideally Phase 14 identifies this need itself. #feature #supervisor #self-improvement #auto-dispatch ~4h model:sonnet blocked-by:t1085.5 ref:GH#1618 assignee:marcusquinn started:2026-02-18T02:11:21Z logged:2026-02-18
 - [ ] t1081 Daily skill auto-update pipeline — add daily skill freshness check to auto-update-helper.sh (24h gate), call skill-update-helper.sh --auto-update --quiet to pull upstream changes for all imported skills. Layer 1: all users get fresh skill docs locally within 24h. Layer 2 (t1082): maintainer PR pipeline for repo-level updates. #enhancement #skills #auto-dispatch ~3h (ai:2h test:1h) model:sonnet ref:GH#1577 logged:2026-02-17
-  - [ ] t1081.1 Add daily skill check to auto-update-helper.sh cmd_check() — after existing version check + setup.sh, call skill-update-helper.sh --auto-update --quiet. Gate with last_skill_check timestamp in state file, skip if <24h. #auto-dispatch ~1h model:sonnet ref:GH#1582 blocked-by:t1081.2 assignee:marcusquinn started:2026-02-18T01:09:53Z
+  - [x] t1081.1 Add daily skill check to auto-update-helper.sh cmd_check() — after existing version check + setup.sh, call skill-update-helper.sh --auto-update --quiet. Gate with last_skill_check timestamp in state file, skip if <24h. #auto-dispatch ~1h model:sonnet ref:GH#1582 blocked-by:t1081.2 assignee:marcusquinn started:2026-02-18T01:09:53Z pr:#1591 completed:2026-02-18
   - [ ] t1081.2 Add --non-interactive support to skill-update-helper.sh — ensure --auto-update --quiet works headlessly without prompts, handles errors gracefully, logs to auto-update.log. #auto-dispatch ~30m model:sonnet ref:GH#1583 assignee:marcusquinn started:2026-02-18T02:28:24Z
   - [ ] t1081.3 Update auto-update state file schema — add last_skill_check, skill_updates_applied fields to auto-update-state.json. #auto-dispatch ~15m model:sonnet ref:GH#1584 blocked-by:t1081.1
   - [ ] t1081.4 Update AGENTS.md and auto-update docs — document daily skill refresh behaviour, clarify that repo version wins on next aidevops update. #auto-dispatch ~15m model:sonnet ref:GH#1585 blocked-by:t1081.1
@@ -101,10 +99,7 @@ Tasks with no open blockers - ready to work on. Use `/ready` to refresh this lis
 - [x] t1060 worktree-helper.sh detect stale remote branches before creating new local branch — check `refs/remotes/origin/$branch` before creating, warn user if stale remote exists, offer to delete if merged. #enhancement #git #workflow #auto-dispatch ~30m model:sonnet ref:GH#1506 assignee:marcusquinn started:2026-02-17T18:33:09Z logged:2026-02-16 pr:#1567 completed:2026-02-17
 - [x] t1077 Fix ShellCheck SC2034 warnings — remove unused variables across 9 files (30 findings): tech-stack-helper.sh (9), supervisor-helper.sh (6), sonarcloud-autofix.sh (4), supervisor/deploy.sh (3), supervisor/dispatch.sh (2), coderabbit-cli.sh (2), code-audit-helper.sh (2 SC2221/SC2222 pattern override), setup/_backup.sh (1), test-orphan-cleanup.sh (1). Also fix SC2068 in tech-stack-helper.sh:2417 (double-quote array expansion). #bugfix #quality #shellcheck #auto-dispatch ~30m model:sonnet ref:GH#1571 assignee:marcusquinn logged:2026-02-17 pr:#1576 completed:2026-02-17
 - [x] t1078 Add explicit return statements to 21 shell scripts missing them — SonarCloud S7682 violations. Files: cron-dispatch.sh, list-verify-helper.sh, session-distill-helper.sh, setup/_backup.sh, setup/_bootstrap.sh, setup/_deployment.sh, setup/_installation.sh, setup/_migration.sh, setup/_opencode.sh, setup/_services.sh, setup/_shell.sh, setup/_tools.sh, setup/_validation.sh, show-plan-helper.sh, subagent-index-helper.sh, supervisor/_common.sh, supervisor/git-ops.sh, supervisor/lifecycle.sh, test-orphan-cleanup.sh, test-pr-task-check.sh, test-task-id-collision.sh. Add `return 0` at end of each function. #bugfix #quality #sonarcloud #auto-dispatch ~30m model:sonnet ref:GH#1571 assignee:marcusquinn started:2026-02-17T19:22:25Z logged:2026-02-17 pr:#1575 completed:2026-02-17
-<<<<<<< Updated upstream
 - [ ] t1079 Add `set -euo pipefail` error handling to scripts missing it — CodeRabbit audit found 8 scripts without standard error handling: cloudron-package-helper.sh, credential-helper.sh, crewai-helper.sh, langflow-helper.sh, autogen-helper.sh, add-skill-helper.sh, agent-test-helper.sh, auto-update-helper.sh. Add `set -euo pipefail` after shebang, verify no breakage from unbound variables or pipefail. #bugfix #quality #coderabbit #auto-dispatch ~30m model:sonnet ref:GH#1572 assignee:marcusquinn logged:2026-02-17
-=======
->>>>>>> Stashed changes
 - [x] t1080 Delete archived scripts in `.agents/scripts/_archive/` — 17 obsolete scripts (add-missing-returns.sh, fix-shellcheck-critical.sh, fix-auth-headers.sh, fix-common-strings.sh, comprehensive-quality-fix.sh, etc.) that are no longer needed. Remove the directory entirely if empty after deletion. #chore #cleanup #auto-dispatch ~15m model:sonnet ref:GH#1573 assignee:marcusquinn logged:2026-02-17 pr:#1574 completed:2026-02-17
 - [x] t1075 Fix verification infinite loop, stuck verifying recovery, and bot review dismissal — three systemic supervisor fixes: (1) cap verify_failed→deployed retries at 3 via state_log counting to prevent infinite loop, (2) recover tasks stuck in 'verifying' for >5min by resetting to deployed, (3) fix dismiss_bot_reviews() regex to match 'coderabbitai[bot]' login format. Also fix tech-stack-helper.sh unused constant and add missing subagent-index entries. #bugfix #supervisor #auto-dispatch ~30m ref:GH#1565 assignee:marcusquinn started:2026-02-17T14:15:00Z logged:2026-02-17 pr:#1566 completed:2026-02-17
 - [x] t1063 Tech stack lookup orchestrator agent and command — create `tools/research/tech-stack-lookup.md` agent and `tech-stack-helper.sh` CLI that orchestrates multiple open-source tech detection tools to replicate BuiltWith.com capabilities. Two modes: (1) **single-site lookup** — detect the full tech stack of a given URL (frameworks, CMS, analytics, CDN, hosting, bundlers, UI libs, state management, etc.), (2) **reverse lookup** — find sites/apps using specific technologies with filtering by region, industry, keywords, traffic tier. The orchestrator calls provider subagents (t1064-t1067) in parallel, merges/deduplicates results into a unified report, and caches results in SQLite. Output formats: terminal table, JSON, markdown report. Add `/tech-stack <url>` slash command. Update subagent-index and AGENTS.md progressive disclosure table. #feature #agent #research #auto-dispatch ~4h (ai:3h test:1h) ref:GH#1521 assignee:marcusquinn logged:2026-02-16 pr:#1541 completed:2026-02-17