sybra

Review what shipped in the OpenAI Codex CLI since the last adoption analysis and file an umbrella issue (☂️) + one subissue per adoptable feature, each grounded in how Sybra actually drives Codex headless. Use when asked to "check what's new in Codex", "adopt new Codex features", or to refresh the Codex CLI adoption tracker. For Claude Code, use claude-feature-adoption instead.

Fast adversarial code review. Red-teams a diff or PR to find the concrete bug, then refutes its own findings to cut false positives. Works with Claude, Codex, or Copilot — uses subagents when the runtime has them, falls back to sequential local passes. Use for routine changes where a full staff review is overkill.

Review what shipped in the OpenAI Codex CLI since the last adoption analysis and file an umbrella issue (☂️) + one subissue per adoptable feature, each grounded in how Sybra actually drives Codex headless. Use when asked to "check what's new in Codex", "adopt new Codex features", or to refresh the Codex CLI adoption tracker. For Claude Code, use claude-feature-adoption instead.

Non-interactive PR review fix — fetch unresolved review threads, auto-apply valid fixes, reply on every thread (applied/questionable/invalid). Use when given a PR URL and asked to address review feedback without prompts.

Analyze Sybra health findings + audit summary to explain workflow issues and propose grounded fixes. Use when asked to review how work is flowing or suggest process improvements.

Analyze the Sybra evaluation scorecard, breakdowns, and weaknesses, then recommend concrete improvements toward an autonomous mid-level engineering team. Use when asked how well/efficiently Sybra is working, to review the evaluation data, or to act on the fleet scorecard.

Evaluate completed Sybra tasks — determine status transition and link PRs. Use when asked to evaluate task completion.

Hand a researched, already-decided task off to Sybra for autonomous implementation or cross-provider review/testing. Use after you have explored a problem in an Orca (or any) git worktree, agreed on the approach, and want Sybra to skip its own planning and start at the right stage — reusing this exact worktree. Triggers on "hand this off to Sybra", "let Sybra implement this", "ship this to Sybra", "handoff to sybra". Works under Claude Code, Codex, and Copilot.

Deprecated. Monitor now runs in-process inside Sybra. Use `sybra-cli monitor scan` for an ad-hoc detector pass or read the `monitor:report` event from the GUI.

Interactive operator planning for a Sybra task when a human explicitly asks to plan or replan in chat. Do not use for workflow-dispatched plan/review/test/implementation agents or prompts that already contain stage instructions.

Periodic deep analysis of agent runs and workflow health. Reads structured findings from Go health checker, investigates root causes by reading agent logs, cross-correlates patterns, and files deduped GitHub issues. Designed for loop-agent invocation every ~6h.

Interactive operator task CRUD for Sybra when a human explicitly asks to list, create, update, or delete tasks in chat. Do not use for workflow-dispatched plan/review/test/implementation agents or prompts that already specify task-stage instructions.

Adversarially test a Sybra task in the testing phase. Start the real app/cluster in the task's isolated sandbox/worktree and try to PROVE the implementation does NOT satisfy the task description — exercising the happy path, edge cases, boundaries, and abuse. Use when a task enters the testing status, or when asked to "test this feature", "verify the change works", "prove it works", "run manual tests", "break it". Plans nothing and reviews no test plan — it executes.

Triage Sybra tasks — delegate to the Go classifier which rewrites the title, assigns tags/mode/status, and matches a project in one atomic update. Use when asked to triage, categorize, or prioritize tasks.

Diagnose why Sybra tasks are stuck in human-required, find the root cause, file a GitHub issue (or a scrubbed local task for work-typed projects) documenting it, and unblock the task back into the workflow. Use when asked to "check why tasks need human", "find root cause of stuck tasks", "unblock human-required tasks", or "file an issue and get this task moving again". Complements — does not duplicate — the automated human-review agent that already fires on the human-required transition. Works under Claude Code, Codex, and Copilot — every step is a plain `sybra-cli`/`gh`/`git` Bash call, no provider-specific tooling.

Non-interactive PR review fix — fetch unresolved review threads, auto-apply valid fixes, reply on every thread (applied/questionable/invalid). Use when given a PR URL and asked to address review feedback without prompts.

Critique an implementation plan before approving execution. Spawns persona subagents when available, falls back to local review, auto-applies refinements on REFINE verdict, and saves every review to ~/.claude/plan-reviews/. Use when a plan has been generated (Plan Mode output, ExitPlanMode, pasted plan text, or plan file path) and needs review before code is written.

Review what shipped in Claude Code since the last adoption analysis and file an umbrella issue (☂️) + one subissue per adoptable feature, each grounded in how Sybra actually drives Claude headless. Use when asked to "check what's new in Claude Code", "adopt new Claude features", or to refresh the Claude Code adoption tracker. For the Codex CLI, use codex-feature-adoption instead.

Merge two independent implementation plan drafts into one clean Sybra plan. Use when given two plan file paths and an optional --task/--out.