TL;DR —codeep --yolo -- "<prompt>"starts Codeep for a launcher with nothing left to answer, Omarchy's skills in~/.agents/skillsload, and Codeep's colours follow the Omarchy theme.
Added
codeep -- <prompt>starts a new session with<prompt>as its first
message (also -p <prompt> and --prompt <prompt>). Everything after -- is the prompt, so codeep -- review asks the model and does not run codeep review, and a leading / is not run as a slash command either: codeep -- /commit sends "/commit" to the model. A missing or doubled prompt is an error rather than a guess.
codeep --yolostarts without stopping to ask, for this launch only:
agent actions run without confirmation, a vague task gets no clarifying questions, the folder is used as the project with read & write access, and a new session starts. Nothing is saved, so the next plain codeep asks as before; changing Agent Confirmation or Agent Interactive Mode in /settings ends --yolo's hold on that setting, and so does loading a saved profile for Agent Confirmation. Writes to files that decide what runs later are still confirmed, and a workspace's own MCP servers still ask to be trusted. The status bar shows YOLO while confirmations are off.
- Skills in
~/.agents/skillsload, the directory agent tools share and
where Omarchy links its skills. When two skills have the same name, the project's .codeep/skills wins, then ~/.codeep/skills, then ~/.agents/skills. /skills bundles names the directory each came from.
- A skill can send the agent to the files beside it.
invoke_skillnow
gives the skill's directory and reads a file from it (a guide such as reporting.md), which read_file cannot reach for a skill outside the project. Only files inside that directory are read.
- Codeep follows the Omarchy theme. On Omarchy, the colours Codeep
paints itself — the brand red, the greys of the chat, the code colours, the warning orange and the YOLO badge — come from the current theme's colors.toml and change when the theme does, in a session that is already open, the way Claude Code, OpenCode and Pi do. A theme colour too faint to read as text is darkened or lightened until it reads, so light themes stay legible. Follow Omarchy theme in /settings (On by default, shown only on Omarchy) turns it off. Codeep only reads Omarchy's state; it writes nothing to your Omarchy config. Elsewhere every colour is exactly what it was.
Fixed
- Startup asks its questions one at a time. In a folder that is not a
repository but has its own MCP servers, the question whether to trust them could replace "Set as Project?", leaving that one unanswered and the startup stopped. It now comes after the startup questions, and a launch prompt is sent after it.
- Folded and literal descriptions. A
SKILL.mdwhosedescription:was
written as a > or | block, as Omarchy's are, loaded with ">" as its description.
- The agent sees a skill's whole description. The catalog cut every entry
at 200 characters, mid-word, which dropped the part of a long description that says when to use the skill.
Security
- Writing to
.agents/skills/is confirmed in every mode, as
.codeep/skills/ already was: every Codeep session loads the skills in ~/.agents/skills.