using-taskferry

A guide for sending long-running coding tasks to background agents through the Taskferry command-line tool. Each task runs in an isolated copy and produces a change set for review before acceptance.

In plain words
What is it for?
Use it to decide when to delegate bugs, fixes, or backlog items, choose a worker model, monitor progress, and review or accept the result.
Why use it?
It separates lengthy work from the main session while keeping the resulting code changes inspectable and rejectable.

Skill for Claude CodeCodex

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add skills/jeremysball/taskferry/using-taskferry
Any agent
npx skills add jeremysball/taskferry --skill using-taskferry
Clone the repo
git clone --depth 1 https://github.com/jeremysball/taskferry

Made for: Claude Code, Codex.

Per session 21 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 9,369 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00021 $0.09369
Opus 5 $0.00010 $0.04685
Sonnet 5 $0.00004 $0.01874
Haiku 4.5 $0.00002 $0.00937

Measured yesterday against content hash 8f04dbeaa2fe, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

using-taskferry scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured yesterday.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

integrations/claude/skills/using-taskferry/SKILL.md · 702 lines

How it starts

The opening of the file, as written. The whole thing — 702 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Taskferry Worker Backend

Taskferry dispatches long-running background agents — ferries — through the taskferry CLI, on an external worker model rather than the host runtime. Each dispatch runs in a copy-on-write overlay over a directory you name, and settles into a changeset you accept or reject, so a ferry always produces a verifiable diff before anything lands.

This file is the core loop: decide, dispatch, wait, verify, accept. Four resources hold the detail:

Resource Read it when
resources/deciding-to-dispatch.md Before dispatching at all — the gate that decides whether this is ferry work
resources/choosing-a-model.md Picking --model / --variant: tiers, role mapping, panels
resources/monitoring-progress.md Waiting on a run, relaying progress, arming a fleet watch
resources/failure-modes.md A dispatch crashed, lied, or silently answered the wrong question

Sizing The Task Before Dispatching

See resources/deciding-to-dispatch.md for the gate and self-check before routing a backlog item, bug, or fix through Taskferry — worktree creation, a written brief, dispatch, wait, review are all costs to weigh against a one-or-two-call Read/Edit/Grep. That resource's gate applies unchanged here; this file assumes you've already passed it.

Always Use A Worktree

Every sandboxed ferry writes to a copy-on-write overlay, never the real directory directly -- a rogue or mistaken dispatch cannot corrupt whatever directory you point it at, worktree or not. taskferry result --diff also handles HEAD drift safely rather than silently corrupting: if the directory's real git HEAD has moved since dispatch (someone or something checked out a different branch there while the task was in flight), extraction reports a headDrift field and tries to recover via git apply --3way against the new HEAD. If recovery succeeds, the diff is applied and the task settles pending for normal accept/reject; if the 3-way merge fails, the task is auto-rejected with a changesetError; if recovery couldn't be determined, the task settles pending with changesetError for you to look at by hand. The diff is never silently computed against the wrong tree, but extraction no longer refuses outright — it always records something (taskferry#261 replaced the old fail-closed "HEAD moved" refusal with this recover-or-flag behavior).

Read the full file on GitHub · 702 lines

Files

What ships with it

4 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. yesterday First seen · 702 lines · 21 tokens per session scan A 8f04dbeaa2fe

Subscribe to this mod's changes

using-taskferry is a skill published in the GitHub repository jeremysball/taskferry (0 stars, last pushed yesterday), licensed MIT. It adds 21 tokens to every session and 9,369 once invoked, about $0.0001 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.