AGENT SKILL

delegate-context

The delegate kit: a harness-neutral delegation engine, the agent-definitions renderer and validator, and the delegation skills, in Python with a pytest suite. Use when you work in delegate and you need one of its reference documents: `docs/agents/delegation.md`, `AGENTS.md`.

License 许可
MIT
最近更新
2026-11-15
Stars
★ 0
Forks
0

01 前言 Front Matter

name
delegate-context
description
The delegate kit: a harness-neutral delegation engine, the agent-definitions renderer and validator, and the delegation skills, in Python with a pytest suite. Use when you work in delegate and you need one of its reference documents: `docs/agents/delegation.md`, `AGENTS.md`.

02 资源文件 1

03 SKILL.md 正文

<!-- Generated by agent-definitions bootstrap. Do not change this file by hand; the next run replaces it. Re-run from the repository root:

 agent-definitions bootstrap --repo . --name delegate --domain 'The delegate kit: a harness-neutral delegation engine, the agent-definitions renderer and validator, and the delegation skills, in Python with a pytest suite.' --tier strong --skill tdd --skill test-quality --skill diagnosing-bugs --reference '`docs/agents/delegation.md` — the gates and the hotspots' --reference '`AGENTS.md` — the layout of the repository' --prompt-file .git/delegate/agents/prompt.md --max-turns 150 --gate-command 'nix develop -c python3 -m pytest' --gate-command 'nix flake check' --gate-command uvx --gate-command 'uvx "zensical==$(cat .zensical-version)" build --strict' --gate-command 'claude plugin validate' --gate-command 'claude plugin test' --get-only-command 'gh issue view' --get-only-command 'gh issue list' --output-language ste --skills-root .claude/skills

-->

delegate reference material

The delegate kit: a harness-neutral delegation engine, the agent-definitions renderer and validator, and the delegation skills, in Python with a pytest suite.

This skill holds no copy of the documents below. Each block runs a command that reads the repository, so what you see is the branch you work on, not a snapshot.

Each block is one plain git grep command. Its :/ pathspec names the file from the top of the repository, so the block works from the repository root and from each subdirectory, and in a linked worktree it reads the files of that worktree. Claude Code runs a load-time block only when it can analyse the command, and a block with $(...), ;, if or a pipe stops a subagent from starting. No Claude Code documentation states this rule. It is a measured behaviour: on Claude Code 2.1.283, on 2026-09-26, a compound block stopped a subagent from starting in the manual, acceptEdits and auto permission modes, and one plain command ran. The measurements are in ADR 0006 of the delegate kit, docs/adr/0006-load-time-blocks-are-one-plain-command.md in the repository of the kit. If a block returns an error, your session is not in the repository. If a block returns no output, its reference file can also be missing or renamed. Each path in this skill is relative to the repository root, and git rev-parse --show-toplevel prints that root.

A document of at most 200 lines, counted when this skill was written, is printed in full. A longer document gives its heading index and the command that reads one section of it.

docs/agents/delegation.md — the gates and the hotspots

The block prints this document in full, up to 200 lines.

git grep --untracked --no-exclude-standard -h -E -m 200 -e '^' -- :/docs/agents/delegation.md

AGENTS.md — the layout of the repository

The block prints this document in full, up to 200 lines.

git grep --untracked --no-exclude-standard -h -E -m 200 -e '^' -- :/AGENTS.md

04 同仓库更多 Skills