AI Agent Workflow Collection

6-phase directive-emitter pipeline for /uberdev:cluster — preflight → fetch → chunk → analyze (parallel Task fanout) → propose → execute (--execute only).

Internal 4-phase pre-flight/plan/merge-resolve/sync pipeline for the /merge command. Invoked exclusively by commands/merge.md; do not call directly.

Shared post-implementation review fanout — dispatches 6 advisory reviewer agents (code-reviewer, silent-failure-hunter, type-design-analyzer, comment-analyzer, pr-test-analyzer, plus one general code-quality reviewer) IN A SINGLE MESSAGE and aggregates findings. Use exclusively from /uberdev:review-pr Phase 1, after PR push. Pre-push call sites in /solve and subagent-driven-dev have been retired.

Shared fixed-fleet Workflow engine for /uberdev:uberscan (read-only audit) and /uberdev:ubersimplify (preserve-behavior refactor). Not invoked directly — uberscan-pipeline (mode=scan) and ubersimplify-pipeline (mode=simplify) emit args and mandate the Workflow call into skills/scan-fleet/workflow.js. RFC 0012 §3.7 (Phase 3).

Shared launcher pipeline for /uberdev:solve and /uberdev:turbo. Parses arguments, classifies tier, writes tier-appropriate prompt, dispatches an autonomous agent through the resolved backend per issue. The executable lives in lib/solve-launcher.sh and runs as ONE Bash call; this skill is the contract + triage reference. Invoked by both commands — never call directly.

Use when executing implementation plans with independent tasks in the current…

Use when /uberdev:testers is invoked. Orchestrates a read-only 8-agent adversarial QA audit squad (6 personas + 2 monitors) across N coordinated rounds against a web/api/native target via an on-disk Workflow script; routes findings into findings-to-issues. RFC 0012 §3.10 proving-ground migration.

Use when the user wants to repo-wide issue similarity analyzer and fold-into-lead consolidator. Default --dry-run; --execute mutates. Invokable explicitly as $uberdev-cmd-cluster. Original description: Repo-wide issue similarity analyzer and fold-into-lead consolidator. Default --dry-run; --execute mutates.

Use when the user wants to fast lane for a minimal working prototype or small function. Decomposes the idea, builds it via parallel subagents in-session, runs one light review, opens a PR labelled prototype, and auto-files a harden issue. Skips spec/plan and full /review-pr by design. Invokable explicitly as $uberdev-cmd-dev. Original description: Fast lane for a minimal working prototype or small function. Decomposes the idea, builds it via parallel subagents in-session, runs one light review, opens a PR labelled prototype, and auto-files a harden issue. Skips spec/plan and full /review-pr by design.

Use when the user wants to autonomous loop that dispatches /uberdev:orchestrator (solve + push PR) → auto /review-pr (trust-signal) → /merge if GREEN, recursing on BLOCKER/CRITICAL review-pr-finding issues until convergence or circuit-breaker halt. Inherits the dispatch backend resolver from /solve and /turbo. Inside /goal only: auto-chain to /merge is allowed (carve-out from feedbackmergeindependent). Invokable explicitly as $uberdev-cmd-goal. Original description: Autonomous loop that dispatches /uberdev:orchestrator (solve + push PR) → auto /review-pr (trust-signal) → /merge if GREEN, recursing on BLOCKER/CRITICAL review-pr-finding issues until convergence or circuit-breaker halt. Inherits the dispatch backend resolver from /solve and /turbo. Inside /goal only: auto-chain to /merge is allowed (carve-out from feedbackmergeindependent).

Use when the user wants to create a GitHub issue with classification, codebase investigation, duplicate search, and label/scope validation. Invokable explicitly as $uberdev-cmd-issue.

Use when the user wants to after a PR is approved, lands it into the integration branch — ordering, strategy, conflict resolution, and local sync automated. Invokable explicitly as $uberdev-cmd-merge.

Use when the user wants to comprehensive PR review using specialized agents. Invokable explicitly as $uberdev-cmd-review-pr.

Use when the user wants to review changed code for reuse, quality, and efficiency, then fix any issues found. Invokable explicitly as $uberdev-cmd-simplify.

Use when the user wants to spawn an autonomous agent per GitHub issue via a platform-aware dispatch backend (claude-bg / wezterm / background / codex; auto-selected per host), with auto-triage and tier-appropriate workflow. Monitor via claude agents (claude-bg), visible panes (wezterm), or PID/log/result files (background/codex). Accepts multiple issue numbers; dispatches one agent per issue. Override the backend with --backend= . Invokable explicitly as $uberdev-cmd-solve. Original description: Spawn an autonomous agent per GitHub issue via a platform-aware dispatch backend (claude-bg / wezterm / background / codex; auto-selected per host), with auto-triage and tier-appropriate workflow. Monitor via claude agents (claude-bg), visible panes (wezterm), or PID/log/result files (background/codex). Accepts multiple issue numbers; dispatches one agent per issue. Override the backend with --backend= .

Use when the user wants to read-only adversarial QA audit squad. Spawns 8 agents (6 personas + 2 monitors) over 3 coordinated waves against an auto-detected target (web/api/native/all). Findings are evidence-anchored against a 10-invariant oracle library and auto-filed as GitHub issues via findings-to-issues. The squad never writes app code. Invokable explicitly as $uberdev-cmd-testers. Original description: Read-only adversarial QA audit squad. Spawns 8 agents (6 personas + 2 monitors) over 3 coordinated waves against an auto-detected target (web/api/native/all). Findings are evidence-anchored against a 10-invariant oracle library and auto-filed as GitHub issues via findings-to-issues. The squad never writes app code.

Use when the user wants to unattended /solve — auto-accepts brainstorm recommendations for medium/large issues. Dispatches N parallel autonomous agents via a platform-aware dispatch backend (claude-bg / wezterm / background / codex; cap: 6 default, configurable via fanoutconcurrency.solvebg). Monitor with claude agents (claude-bg), visible panes (wezterm), or PID/log/result files (background/codex). Override the backend with --backend= . Invokable explicitly as $uberdev-cmd-turbo. Original description: Unattended /solve — auto-accepts brainstorm recommendations for medium/large issues. Dispatches N parallel autonomous agents via a platform-aware dispatch backend (claude-bg / wezterm / background / codex; cap: 6 default, configurable via fanoutconcurrency.solvebg). Monitor with claude agents (claude-bg), visible panes (wezterm), or PID/log/result files (background/codex). Override the backend with --backend= .

Use when the user wants to whole-codebase read-only audit. Packs the repo into a fixed fleet of byte-balanced areas (default 8) and runs ONE multi-lens reviewer per area, plus an inline repo-global Semgrep + test-coverage pass. Aggregates findings into a markdown report and files deduped GitHub issues. Never writes code. Whole-repo by default; pass a path/glob to narrow. Invokable explicitly as $uberdev-cmd-uberscan. Original description: Whole-codebase read-only audit. Packs the repo into a fixed fleet of byte-balanced areas (default 8) and runs ONE multi-lens reviewer per area, plus an inline repo-global Semgrep + test-coverage pass. Aggregates findings into a markdown report and files deduped GitHub issues. Never writes code. Whole-repo by default; pass a path/glob to narrow.

Use when the user wants to whole-codebase 3-lens simplification (Reuse/Quality/Efficiency). Packs the repo into a fixed fleet of byte-balanced areas (default 8), audits each area with ONE multi-lens code-simplifier, applies preserve-behavior fixes via code-fixer as one refactor: commit per area on a new branch, opens ONE PR for review, and files leftover blocker findings as GitHub issues. Whole-repo by default; pass a path/glob to narrow. --audit-only for a read-only scan. Invokable explicitly as $uberdev-cmd-ubersimplify. Original description: Whole-codebase 3-lens simplification (Reuse/Quality/Efficiency). Packs the repo into a fixed fleet of byte-balanced areas (default 8), audits each area with ONE multi-lens code-simplifier, applies preserve-behavior fixes via code-fixer as one refactor: commit per area on a new branch, opens ONE PR for review, and files leftover blocker findings as GitHub issues. Whole-repo by default; pass a path/glob to narrow. --audit-only for a read-only scan.

Use when the user wants to read-only cross-domain ideation engine. Fans an agent fleet over parallel islands to generate, gap-fill, combine, and falsify novel approaches across donor domains (SWE/gamedev/IT/math + far-field wildcards). Emits a ranked dossier with a moonshot lane and files top ideas as GitHub issues. Invokable explicitly as $uberdev-cmd-uberthink. Original description: Read-only cross-domain ideation engine. Fans an agent fleet over parallel islands to generate, gap-fill, combine, and falsify novel approaches across donor domains (SWE/gamedev/IT/math + far-field wildcards). Emits a ranked dossier with a moonshot lane and files top ideas as GitHub issues.

Use when /uberdev:uberscan is invoked. Read-only whole-codebase audit — packs the repo into a fixed fleet of byte-balanced areas (default 8), runs ONE multi-lens reviewer per area, runs an inline repo-global Semgrep SAST + test-coverage pass, aggregates findings into a markdown report, and files deduped GitHub issues. Never writes source code. The orchestration runs in the shared scan-fleet Workflow (skills/scan-fleet/workflow.js, mode=scan); this skill is the thin preflight + args seam.

Use when /uberdev:ubersimplify is invoked. Whole-codebase 3-lens simplification — packs the repo into a fixed fleet of byte-balanced areas (default 8, shared lib/chunk.py), audits each area with ONE multi-lens code-simplifier (Reuse/Quality/Efficiency), applies preserve-behavior fixes via code-fixer as one refactor: commit per area on a shared branch, opens ONE PR, and files leftover blocker findings as GitHub issues. --audit-only is read-only. The orchestration runs in the shared scan-fleet Workflow (skills/scan-fleet/workflow.js, mode=simplify); this skill is the thin preflight + args seam.

7-wave island-aware directive-emitter for /uberthink — frames the goal, runs N parallel evolutionary islands (diverge → gap-gate → combine → converge → falsify → genetic-loop-back ≤3) → cross-pollinate → rank → deliver dossier + file top…

Use when starting any conversation - establishes how to find and use skills, requiring Skill tool invocation before ANY response including clarifying…