prath-mode — Claude Code skill: what it does, how to install
Routes user goals to the right skill or execution playbook. It manages multi-step engineering workflows from design checks to pull requests.
npx skills add prathamdby/skills --skill prath-mode -a claude-codeWhat prath-mode does
The skill directs incoming tasks into predefined playbooks. Routes include phased design, diff cleanup (deslop), commit generation, and PR management. An immediate single action takes priority over a full workflow chain.
Work happens in two clear stages: route matching and step execution. It copies playbook steps directly into the agent checklist and verifies that required leaf skills exist on disk. If a referenced SKILL.md is missing, the run stops immediately and asks for npx skills@latest add prathamdby/skills.
Delivery chains require a clean Git working tree before opening a PR. The process halts whenever a step reports an awaiting-user or blocked state.
What it helps with
- route complex goals to specific execution playbooks
- run complete delivery workflows from design to pull request
- enforce diff cleanup prior to committing changes
- verify required playbook files before starting an execution chain
Example prompts for the agent
- Design this feature and run the ship workflow to open a PR
- Clean up recent diffs and commit current changes
- Map the repository structure using the understand-repo playbook
How to install prath-mode
1Skills CLI (Vercel)
npx skills add prathamdby/skills --skill prath-mode -a claude-code2Manually: put the files in place
~/.claude/skills/prath-mode/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/prath-mode/
1Skills CLI (Vercel)
npx skills add prathamdby/skills --skill prath-mode -a codex2Manually: put the files in place
.agents/skills/prath-mode/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add prathamdby/skills --skill prath-mode -a cursor2Manually: put the files in place
.cursor/skills/prath-mode/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install prathamdby/skills prath-mode2Manually: put the files in place
.github/skills/prath-mode/SKILL.md1Skills CLI (Vercel)
npx skills add prathamdby/skills --skill prath-mode -a gemini-cli2Platform CLI
gemini skills install https://github.com/prathamdby/skills --path skills/engineering/prath-modeManually: put the files in place
.devin/skills/prath-mode/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add prathamdby/skills --skill prath-mode -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about prath-mode
How to install prath-mode in Claude Code?
Run npx skills add prathamdby/skills --skill prath-mode -a claude-code. The file will be placed at ~/.claude/skills/prath-mode/SKILL.md.
What does prath-mode do?
It acts as a router that maps user goals to specific playbooks, creates a verifiable task checklist, and tracks execution until completion.
What happens when required skills are missing?
It verifies leaf skill paths before running any step. If a target file is absent, it stops and prompts you to install dependencies with npx skills@latest add prathamdby/skills.
Source text
SKILL.md · MIT
--- name: prath-mode description: Route a goal to a skill or delivery chain. disable-model-invocation: true --- # Prath mode Leaves own invocation, options, procedures, and terminals. Read the selected leaf; playbooks own predetermined order, not copies of leaf behavior. ## Matchers - `design`: phased design or milestone check, `playbooks/design.md` - `autoplan`: autonomous plan, `playbooks/autoplan.md` - `review-plan`: proposal review, `playbooks/review-plan.md` - `deslop`: diff cleanup, `playbooks/deslop.md` - `commit`: commit only, `playbooks/commit.md` - `open-pr`: publish/update PR, `playbooks/open-pr.md` - `finish-pr`: feedback fixes, `playbooks/finish-pr.md` - `inspect-pr`: PR/CI inspection or reply, `playbooks/inspect-pr.md` - `understand-repo`: map/read saved map, `playbooks/understand-repo.md` - `research-external`: local external source, `playbooks/research-external.md` - `orchestrate`: delegated work, `playbooks/orchestrate.md` - `session`: handoff/resume, `playbooks/session.md` - `cursor-agent`: Cursor CLI, `playbooks/cursor-agent.md` - `claude-code`: Claude CLI, `playbooks/claude-code.md` - `codex`: Codex CLI, `playbooks/codex.md` - `todoist`: task creation/preview, `playbooks/todoist.md` - `use-skill`: remote skill execution, `playbooks/use-skill.md` - `restate`: plain restatement and catch-up, `playbooks/restate.md` - `ship`: implement/deslop/commit/PR, `playbooks/ship.md` - `design-then-ship`: design and delivery, `playbooks/design-then-ship.md` - `save-work`: optional cleanup/commit, `playbooks/save-work.md` An immediate action outranks a chain. Review does not prepend to ship. Inspection/reply stays inspect-pr; fixing feedback selects finish-pr. Neither inspect-pr nor gh escalates into fix-pr; fix-pr owns its gh call. Deslop is required in ship, optional in save-work; resolve mixed index/worktree paths before it. Parent owns implementation. ## 1. Match and verify Record `route | owner | completed | design | diff/tests | terminal`. Open one playbook and copy its steps into todos verbatim. Two matching actions/ chains → one clarification; no match → one outcome question, then design if large. From playbooks resolve leaves as `../../<name>/SKILL.md` or `../../../personal/<name>/SKILL.md`; check all before starting and each again before its turn. Missing → name paths, stop, offer `npx skills@latest add prathamdby/skills`. Done when route, completion condition, and installed owners are fixed. ## 2. Execute and resume Read each current leaf in full; run it to a terminal. A reported terminal ends an action route, but does not necessarily succeed. Advance a chain only on the leaf's evidenced success/no-op. Blocked/waiting/awaiting-push pauses regardless of a playbook's terminal-observed completion wording. For upfront-design, DIRECT is no-op; DESIGNED/CHECKED are success; AWAITING_USER/BLOCKED never advance. Parent implementation records planned diff and test evidence. Before make-pr require a clean tree. On interruption revalidate the last owner's artifacts before continuing. Done when the route's successful completion condition holds or a pause is reported with its owner; reporting a blocker is not successful delivery.
© prathamdby · MIT · prathamdby/skills
