AI Agent Workflow Collection

Audit stored normal content on the bound Context Tree's current default branch when a human explicitly asks to audit the whole tree, a domain, or specific normal paths for drift, contradictions, duplication, density, metadata, placement, or relationship problems. Do not use for ordinary task reads, source-backed writes, Context Tree PR/MR reviews, or empty-tree setup.

Review a GitHub pull request or GitLab merge request against the workspace-bound Context Tree when a trusted server-authored Context Reviewer run supplies provider-scoped authority. Repair every safely determined finding with the host git and forge CLI identity, then use GitHub App review plus exact-head merge or GitLab note plus exact-SHA merge. Do not use for code changes, ordinary tree reads or writes, or default-branch audits.

File a GitHub issue about a defect in First Tree itself — the CLI, agent runtime, chat, web app, GitHub integration, GitLab integration, or Context Tree tooling — onto First Tree's own GitHub-hosted public tracker (agent-team-foundation/first-tree). Use only when the user reports that First Tree the platform is broken, errored, crashed, or misbehaving and wants it reported (e.g. "First Tree has a bug", "chat send keeps failing", "the CLI crashed", "给 First Tree 开个 issue 反馈这个 bug"). The skill gathers reproduction steps, the First Tree client/CLI version, the chat ID, the reporting user's ID, OS, and error output, then opens the issue with the user's gh CLI. NOT for filing an issue into the user's own or bound source repo — that is an ordinary source-repo issue using the target forge CLI (gh for GitHub, glab for GitLab), not this skill — and not for bugs in the user's project code or third-party tools.

Act as an independent QA engineer for a software repository. Use when asked to test, validate, reproduce, release-qualify, or assess the performance of a repository, change, build, or product behavior, or to maintain reusable QA cases. Establish a complete runnable QA harness before scoping execution, validate real product behavior, report evidence honestly, and do not modify the product under test.

Read the current repo's Context Tree before acting. Use when the user provides a task, topic, file path, feature name, bug, error, repo area, owner, or other signal and Codex needs to locate and read the relevant context files from the bound context repo. Always inspect first-tree tree tree --help in the context repo first, then use first-tree tree tree filtering options to select candidate files; read the selected file contents with the agent's native file-reading capability.

One-time bootstrap for a brand-new, still-empty Context Tree. Delivers an initial top + second-level domain skeleton drawn from the bound source repos (PR1, user approves), then initial leaf-node content for each approved domain with explicit coverage reporting (PR2). Use right after Cloud onboarding provisions the workspace and tree repo. Refuses on any populated tree. Do not use to write an incremental update from a specific PR / doc / note — that is first-tree-write. Do not use for broad maintenance or drift-audit work on an existing tree.

Use only when a First Tree onboarding start-chat system message explicitly names first-tree-welcome, or for First Tree onboarding welcome, intro, or value-first first-work chats. Do not use for tree setup chats, ordinary chats, PR reviews, tree writes, or maintenance.

Source-driven Context Tree authorship guide. Covers what a Context Tree is, the source-system boundary, authorship read-discipline, and how to write tree updates from a specific source artifact such as a PR, design doc, meeting note, or pasted source. Use first-tree-read for task-scoped tree reads before acting; use this skill when the user asks to reflect a source artifact into the tree, update the tree from source material, or write a decision to the tree. If no source artifact is available, there is no write task; ask the user for one.