Oneshot Ship

oneshot CLI

Ship code with a single command. oneshot is an agentic software workflow runtime: repo + task in, isolated agent run out, reviewed PR ready. It runs with one selected fallback provider, or with invisible adaptive routing when routing.enabled is true. The router chooses Codex or Claude plus reasoning effort, while each provider stays on its configured frontier model. Works over SSH to a remote server or locally with --local.

When to use this skill

• User wants to ship a code change to a repository without manual coding
• User wants to automate the plan/implement/review/PR workflow
• User mentions "oneshot" or wants to delegate a coding task
• User wants to run a task on a remote server or locally

Installation

bun install -g oneshot-ship

Setup

Run oneshot init to configure SSH host, workspace path, API keys, and model preferences. Config is saved to ~/.oneshot/config.json.

Repos on the server should live as <org>/<repo> under the workspace path:

~/projects/
  my-org/my-app/
  my-org/my-api/

Server prerequisites

• Bun
• Either Codex CLI or Claude Code CLI
• GitHub CLI (authenticated)
• OPENAI_API_KEY for Codex mode, or ANTHROPIC_API_KEY for Claude mode

Usage

oneshot <repo> "<task>"                 # ship a task
oneshot <repo> <linear-url>            # ship from a Linear ticket
oneshot <repo> "<task>" --bg           # fire and forget
oneshot <repo> "<task>" --local        # run locally, no SSH
oneshot <repo> "<task>" --mode deep    # skip classification and force deep mode
oneshot <repo> "<task>" --workflow ship # apply a workflow preset
oneshot <repo> "<task>" --deep-review  # force exhaustive review
oneshot <repo> "<task>" --model gpt-5.5 # override configured plan/PR model
oneshot <repo> "<task>" --branch dev   # target a different branch
oneshot <repo> "<task>" --base-path /srv/workspaces  # override repo root for this run
oneshot <repo> "<task>" --worktree-root /tmp/agents   # override temp worktree root
oneshot route "fix failing CI and publish" --json # inspect invisible route
oneshot <repo> --dry-run               # validate only
oneshot init                           # configure
oneshot stats                          # recent runs + timing
oneshot runs                           # durable run ledger
oneshot runs --json --limit 10         # list runs for automation
oneshot status <run-id|events-file> --json  # inspect one run
oneshot eval --json                    # summarize outcomes
oneshot workflow list                  # inspect workflow presets
oneshot workflow show fix-ci --json    # inspect one workflow preset
oneshot policy init                    # create .oneshot/policy.json
oneshot policy init --path ./repo      # write policy in another directory
oneshot mcp serve                      # expose MCP tools over stdio
oneshot doctor                         # package, tool, SSH, and event health
oneshot doctor --repo zkp2p/pay        # health plus checkout existence

Pipeline

1. Validate: checks the repo exists, fetches latest from origin
2. Worktree: creates a temp git worktree from the target base branch
3. Route: oneshot's adaptive router chooses provider, reasoning effort, context shape, execution style, verification profile, and fast/deep mode
4. Plan: routed agent reads the codebase plus AGENTS.md/CLAUDE.md instructions, package scripts, policy packs, and relevant docs, then outputs an implementation plan
5. Execute: routed agent implements the plan. If it times out with partial changes, the pipeline continues
6. Draft PR: configurable agent creates a branch, commits, and writes PR metadata; the runtime pushes and opens or updates the draft PR
7. Review: configurable agent reviews the diff across correctness, compatibility, runtime contracts, policy, security, docs quality, and simplicity. Confirmed issues are fixed directly
8. Finalize: runtime pushes review fixes and marks the PR ready, or preserves the draft if review fails/times out

Every run writes JSONL events to /tmp/oneshot-<runId>.events.jsonl and mirrors them into ~/.oneshot/runs/<runId>.events.jsonl. Worktrees are cleaned up after successful runs and preserved on failure for recovery.

Configuration

~/.oneshot/config.json:

{
  "host": "user@100.x.x.x",
  "basePath": "~/projects",
  "provider": "codex",
  "routing": { "enabled": true },
  "linearApiKey": "lin_api_...",
  "claude": { "model": "opus", "timeoutMinutes": 180 },
  "codex": {
    "model": "gpt-5.5",
    "reasoningEffort": "xhigh",
    "reviewModel": "gpt-5.5",
    "reviewReasoningEffort": "xhigh",
    "timeoutMinutes": 180
  },
  "phases": {
    "classify": { "model": "gpt-5.5", "reasoningEffort": "medium" },
    "plan": { "model": "gpt-5.5", "reasoningEffort": "xhigh" },
    "execute": { "model": "gpt-5.5", "reasoningEffort": "xhigh" },
    "review": { "model": "gpt-5.5", "reasoningEffort": "xhigh" },
    "deepReview": { "model": "gpt-5.5", "reasoningEffort": "xhigh" },
    "pr": { "model": "gpt-5.5", "reasoningEffort": "high" }
  },
  "stepTimeouts": {
    "planMinutes": 20,
    "executeMinutes": 60,
    "reviewMinutes": 20,
    "deepReviewMinutes": 20,
    "prMinutes": 20
  }
}

Only host is required for SSH runs. Local mode works without a config file. Remote SSH runs stream the active oneshot config to the server for that run, so basePath, provider defaults, timeout settings, and configured Linear credentials stay aligned even if the server does not have its own oneshot config file.

Flags

Customization

• Put an AGENTS.md or CLAUDE.md in any repo root. oneshot passes the instruction snapshot to the configured agents, and the prompts ask agents to inspect relevant local instruction files directly
• Choose provider: "codex" or provider: "claude" as the fallback. Set routing.enabled: true when oneshot's adaptive router should silently pick Codex or Claude per task
• Configure claude.model and codex.model as the frontier models for each provider. Adaptive routing varies effort and provider, not model class
• Configure phases.classify, phases.plan, phases.execute, phases.review, phases.deepReview, and phases.pr for model and reasoning defaults when routing is disabled
• Edit prompts/plan.txt, execute.txt, review.txt, pr.txt to change pipeline behavior
• Use oneshot policy init to add .oneshot/policy.json with protected paths, required checks, approval-sensitive keywords, and secret-pattern gates
• Use oneshot mcp serve when another agent should call oneshot through tools instead of shelling out ad hoc
• For dense specs, explainers, review maps, incident reports, design sheets, or one-off editors, oneshot can create a self-contained HTML artifact instead of a long markdown wall. Durable artifacts belong in docs/artifacts/; throwaway local artifacts belong in /tmp/oneshot-html-artifacts/.

Tips

• Use --bg to fire and forget long tasks
• Linear integration moves tickets to "In Review" and comments the PR URL
• Per-step timeouts prevent runaway processes (plan 20m, execute 60m, review 20m, PR 20m)
• Worktree isolation means your main branch is never touched
• Task classification picks fast or deep mode automatically. Use --deep-review to force deep
• Duration estimates come from historical runs per repo (~/.oneshot/history.json)
• Durable run events live in ~/.oneshot/runs; use oneshot runs, oneshot status, and oneshot eval for recovery and feedback loops
• Repo slugs must be exact owner/repo values. Nested paths and .. are rejected before dispatch
• doctor compares the running CLI version with the npm registry, checks config, required CLIs, recent event files, SSH reachability, and can verify a specific checkout with --repo
• oneshot mcp serve exposes tools for running tasks, listing runs, reading status, initializing policy, listing workflows, and summarizing eval outcomes