Agent Claude Code
HarnessFlow worker subagent — spawn only from a HarnessFlow workflow step, never by autonomous delegation. Analyzes codebase architecture for improvement opportunities — identifies inappropriate designs and proposes structural improvements.
Agent Claude Code
HarnessFlow worker subagent — spawn only from a HarnessFlow workflow step, never by autonomous delegation. Analyzes code by following the pipeline diagram from upstream to downstream, ensuring full coverage.
Agent Claude Code
HarnessFlow worker subagent — spawn only from a HarnessFlow workflow step, never by autonomous delegation. Runs target scripts to attempt to reproduce a bug, captures all output, and summarizes the reproduction result for the main agent.
Agent Claude Code
HarnessFlow worker subagent — spawn only from a HarnessFlow workflow step, never by autonomous delegation. Identifies unnecessary complexity and proposes simplifications that preserve behavior while reducing cognitive load.
Agent Claude Code
HarnessFlow worker subagent — spawn only from a HarnessFlow workflow step, never by autonomous delegation. Critically challenges plans and analyses — looks for overlooked side effects, integration risks, incorrect assumptions, or potential regressions. At draft stage, also grills the main agent with up to five pointed…
Agent Claude Code
HarnessFlow worker subagent — spawn only from a HarnessFlow workflow step, never by autonomous delegation. Proposes three to five structurally different alternative plans to the current plan — searching a risky, an aggressive, and a rare archetype — each constraint-checked against the request, each with a calibrated…
Agent Claude Code
HarnessFlow worker subagent — spawn only from a HarnessFlow workflow step, never by autonomous delegation. Executes the planned actions toward a goal based on a finalized plan — validates pre-conditions, runs the actions, captures output, and reports results.
Agent Claude Code
HarnessFlow worker subagent — spawn only from a HarnessFlow workflow step, never by autonomous delegation. Analyzes code and plans by reading only the most directly relevant files, prioritizing depth over breadth.
Agent Claude Code
HarnessFlow worker subagent — spawn only from a HarnessFlow workflow step, never by autonomous delegation. Analyzes code and plans using its own judgment — decides what files to read, in what order, following its own logic.
Agent Claude Code
HarnessFlow worker subagent — spawn only from a HarnessFlow workflow step, never by autonomous delegation. Implements code changes based on a finalized plan — reads associated files, writes code, and reports what was changed.
Agent Claude Code
HarnessFlow worker subagent — spawn only from a HarnessFlow workflow step, never by autonomous delegation. Searches online for resources, tools, packages, patterns, and solutions needed by the workflow.
Agent Claude Code
HarnessFlow worker subagent — spawn only from a HarnessFlow workflow step, never by autonomous delegation. Reviews all plans and comparison statements from a principal engineer perspective — assesses correctness, feasibility, and rejects redundant proposals.
Agent Claude Code
HarnessFlow worker subagent — spawn only from a HarnessFlow workflow step, never by autonomous delegation. Validates implementations from a QA engineer perspective — reviews code changes, runs scripts, and reports correctness.
Agent Claude Code
HarnessFlow worker subagent — spawn only from a HarnessFlow workflow step, never by autonomous delegation. Identifies and proposes removal of redundant code, overlapping implementations, and duplicate functionality.
Agent Claude Code
HarnessFlow worker subagent — spawn only from a HarnessFlow workflow step, never by autonomous delegation. Identifies robustness issues, potential bugs, and proposes improvements to make the codebase more resilient.
Agent Claude Code
HarnessFlow worker subagent — spawn only from a HarnessFlow workflow step, never by autonomous delegation. Reviews plans and code from a senior staff engineer perspective — assesses correctness, feasibility, and potential regressions.
Instructions file CodexOpenCode
AGENTS.md instructions for HangYu8123/HarnessFlow, covering agents.md — codex entry point, activation gate, pack path resolution, workflow execution and repo context files.
Instructions file
Claude Code instructions for HangYu8123/HarnessFlow, covering claude.md — claude code cli entry point, activation gate, pack path resolution, workflow execution and repo context files.
Agent
This file lists all custom agents defined in this pack. Workflow instruction files reference these agents by name when creating subagents.
Skill Claude CodeCodex
Plan and optionally execute a large pull request split into small stacked PRs. Use for PR breakdown, split PR, stacked PRs, branch decomposition, oversized diffs, gh-stack, Graphite, git range-diff, or PR too large.
Skill Claude CodeCodex
Orchestrate Claude Code bundled skills after implementation. Use only when the main agent is Claude Code or another Claude agent with Claude Code skills available and an implementation report exists.
Skill Claude CodeCodex
Multi-axis, review-only code review of a diff across six axes — request achievement, correctness, readability/simplicity, architecture, security, and performance — with severity-labelled findings. TRIGGER: invoked by a HarnessFlow workflow's post-implementation review step when the request header is codereview: local.…
Skill Claude CodeCodex
Simplify code for clarity while preserving exact behavior — reduce nesting, split long functions, remove redundancy and dead code, and fix unclear names. TRIGGER: invoked by a HarnessFlow workflow's post-implementation review step when the request header is simplify: local. Not a planning-time skill.
Skill Claude CodeCodex
Generate MY "last week update report": summarize the past 7 days of git commit history into a short, user-facing weekly update. TRIGGER ONLY when the user explicitly asks to generate their last week / weekly update report (e.g. "generate my last week update report", "make my weekly update", "what did I ship last…