AI Agent Workflow Collection

Use when adding or standardizing project license files and README license sections. Defaults to MIT unless the user or repository explicitly asks for another license, creates/updates a root LICENSE from the bundled template, and appends a bottom-of-README Markdown link such as MIT.

Use when the user asks to implement a plan end-to-end, test it, run codex review against the base branch in a loop until no P0/P1 issues remain, and make sure CI/CD is green before finishing.

Use when asked to summarize what one person or a small set of GitHub identities did over a recent time window, especially daily work reports across org repos. Produces a plain-language activity summary grouped by shipped, closed/replaced, in progress, and planning work, with source links appended to each item.

Use when an agent should open, inspect, or monitor a single GitHub issue or pull request with the ghzinga terminal UI.

Operate and integrate cooperative lifecycle control for detached Hugging Face Jobs with the hf-job-control Python package and CLI, including durable chunked outputs, audits, and incident recovery. Use when creating logical runs, registering immutable launch specifications, launching or monitoring physical Jobs, requesting pause/stop/abort, resuming from verified checkpoints, implementing CheckpointAdapter and Controller boundaries, making batch generation recoverable, running the remote canary, investigating control or checkpoint failures, or proving receipts and provenance for a completed run.

Use when creating, updating, selecting, promoting, auditing, or deleting local runtime environments for inference engines or benchmark toolchains such as vLLM, SGLang, llama.cpp, TensorRT-LLM, or FlashInfer-backed runtimes. Enforces canonical runtime paths, manifests, smoke tests, promotion rules, cleanup rules, and prevents ad hoc virtualenv or toolchain sprawl.

Use when maintaining ~/repos/onurclaw/OPENCLAWONURINVENTORY.md, including periodic roughly every-2-hour refreshes, auditing OpenClaw local model and open-weight model issue/PR inventories, deciding whether a thread belongs in that file, sorting the inventory, updating the reviewed-through issue/PR watermark, or explaining why an item was included or excluded.

Create polished SVG/PNG chart figures from experiment metrics, especially multi-panel bar charts for blog posts or reports. Use when asked to generate, revise, or verify a graph/plot/figure and visual correctness matters.

Use when a prompt appears to be a queued, repeated, resumed, scheduled, or reminder-style instruction and Codex must decide whether to continue unfinished work or ignore it because the requested work is already complete. Implementation-agnostic queue handling for any task type.