Run terminal coding-agent workflows with Phi
Use Phi to run lightweight terminal coding-agent sessions with sub-agents, guarded tool use, hashline edits, and lazy MCP discovery.
npx skills add agentskillexchange/skills --skill run-terminal-coding-agent-workflows-with-phi
Use Phi when an operator wants a lightweight local terminal coding agent that can read, edit, and run repository commands while keeping tool use guarded and context growth under control. The workflow is to configure a model, open the Phi TUI or run a bounded headless prompt, attach local skills and MCP servers only when needed, use sub-agents for isolated jobs, and rely on hashline-anchored edits, permission gates, hooks, and persisted sessions to keep coding work reviewable. Invoke this instead of using a hosted coding product when the user needs a local, inspectable agent harness with permission-gated command execution, reusable skills, persistent sessions, and MCP access that does not dump every tool schema into the model prompt. The scope boundary is operating coding-agent sessions through Phi’s own CLI/TUI runtime; it is not a generic model SDK, a generic MCP server listing, or a broad framework card.
What this skill actually does
Inputs and prerequisites: Phi CLI/TUI, a configured OpenAI-compatible or Anthropic model API key, local git repository, optional MCP servers, optional Phi skills and hooks.
Setup notes: Install from the published GitHub release or build from source with Go 1.26.3+ using git clone https://github.com/pulseaiclub/phi.git, cd phi, and make install. Then run phi config or set PHI_MODEL and PHI_API_KEY, start the TUI with phi, or run a bounded headless task with phi run -p ” “.
Source and verification boundary: use https://pulseaiclub.github.io/docs/getting-started/ as the canonical reference before running the workflow; keep commands, API calls, CLI usage, and generated outputs reviewable against that upstream source.
Framework fit: publish this as a Custom Agents workflow only when the operator can invoke the documented toolchain directly, rather than treating the upstream project as a generic product listing.