You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Commit 334eb9e
Browse filesBrowse the repository at this point in the historyBrowse files
feat(harness): discover agent skills in the sandbox
Add the codework.prompt.skill builtin. Skills are SKILL.md directories
(and standalone .md files in .codework/skills) found through the
session's sandbox filesystem, never the host or <home>:
.codework/skills at the cwd, then .agents/skills from the cwd up to the
git root, or the sandbox root outside a repository. Only name,
description and location go into the <skills> prompt section; the model
reads the file when a task matches. Invalid frontmatter, unreadable
files and name collisions warn instead of failing.
Share the ancestor walk, the batched existence probe and BOM stripping
between the instruction and skill plugins.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Copy file name to clipboardExpand all lines: CHANGELOG.md
+6Lines changed: 6 additions & 0 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -6,6 +6,12 @@ This file is the canonical source for unreleased changes and published release n
6
6
7
7
- Add in-progress work under `## [Unreleased]
8
8
9
+
## [Unreleased]
10
+
11
+
### Added
12
+
13
+
- Added Agent Skills (`codework.prompt.skill`). A skill is a directory with a `SKILL.md` whose frontmatter has a `name` and a `description`; `.codework/skills` may also hold standalone `.md` skills. Skills are found in the session's sandbox, local or remote, and never on the host or in `<home>`: `.codework/skills` in the working directory, then `.agents/skills` in the working directory and each parent up to the git root (or up to the sandbox root outside a repository). `.gitignore`, `.ignore` and `.fdignore` are honoured. The system prompt lists only each skill's name, description and location under `<skills>`; the model reads the full file when a task matches. A skill with `disable-model-invocation: true` is not listed. Invalid fields, unreadable files and name collisions log a warning instead of failing; the first skill found keeps the name. The hint names the `read` tool, or `bash` when `read` is off, and names no tool when the reader is hidden from the prompt. Turn it off with `{ "plugin": "codework.prompt.skill", "enabled": false }` in `plugins`.
0 commit comments