Kimiskill
Delegate tasks from Claude Code to Kimi Code CLI via a structured file-based bridge
npx -y skills add fergus/claude-kimi-delegate --skill kimiskillAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
2 things to look at
- no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
- 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Triggered when Kimi is invoked via the Claude delegation bridge. Use when the task file indicates from: claude, the prompt mentions delegation mode, or the user is acting as a bridge from Claude Code. This skill sets output conventions, prevents clarifying questions, and ensures results are written to the declared output path.
SKILL.md
3.0 KB, 643 tokens by cl100k_base, as published. Nobody here has run it
Claude Delegation Mode
You are being invoked by Claude Code via the delegation bridge. Claude has written a task file and expects a structured result written to a specific file path.
Detection
Delegation context is present when any of the following are true:
- The prompt asks you to read a task file from
.kimi/delegations/ - The prompt mentions "delegation mode" or "Claude delegation"
- A task file with
from: claudefrontmatter is loaded into context
Workflow
- Read the task file if it is provided in the prompt or context
- Extract frontmatter:
task_id,output_path,skill,context_files - Load sub-skill if
skillfield is non-empty (e.g.,skill: ce-plan) - Execute the task following the conventions below
- Write the complete result to
output_path
Conventions
- Be concise. Claude's context window is shared with the rest of the conversation. Avoid verbosity. Aim for under 80 lines in the result file unless the
# Expected Outputsection explicitly asks for depth (e.g., long-form reviews, multi-file plans). - Do not ask clarifying questions. If the task is vague, make one reasonable assumption, execute, and note it under
# Assumptions. - If blocked, fail cleanly. If the task is literally impossible, write
# Blocker: [reason]in the result file and exit. Do not hang or loop. - Use repo-relative paths when referencing files.
- Do not modify the task file.
Output Format
Write the result to the output_path declared in the task file. Use this structure:
# Summary
2-3 sentences describing what was done and the key outcome.
# Details
[Main body of the result. Use sections, lists, and code blocks as appropriate.]
# Assumptions
[Only if you made assumptions due to vague input. List each assumption briefly.]
Edit Tasks vs. Research Tasks
Use the # Expected Output section to decide how deep the result should be:
- Edit / in-place update tasks: If the task asks you to edit an existing file (e.g., "Update
docs/foo.mdin place"), the result file should be lean. Include only:- Which files were modified
- A high-level summary of the changes
- Any assumptions or verification steps Do not duplicate the full edited content into the result file.
- Research / generation tasks: If the task asks for analysis, planning, or new content, the result file should contain the full output.
Sub-Skill Routing
If the task file specifies a skill field:
- Load that skill first
- Follow its workflow and conventions
- Still respect the output path and delegation conventions above
The skill field is optional. When empty, execute as a general delegation.
What ships with it: 1 file
2.5 KB alongside SKILL.md
references/
- delegation-format.md2.5 KB
Gives 0 of the 12 instructions most context ai engineering skills give in 643 tokens
Counted across 1,193 of the 1,976 authors here whose files we hold, read 2026-08-07
- Dispatch a fresh implementer subagent per taskin 48 of 1193, across 19 files
- Dispatch a final code reviewer after all tasksin 33 of 1193, across 8 files
- Provide full task text to the subagentin 30 of 1193, across 9 files
- Review spec compliance before code qualityin 27 of 1193, across 10 files
- Make the hook script executablein 26 of 1193, across 8 files
- Re-snapshot after navigation or DOM changesin 25 of 1193, across 19 files
- Read files before editing themin 22 of 1193, across 11 files
- Answer subagent questions before proceedingin 22 of 1193, across 7 files
- Mark task complete in TodoWrite after approvalin 22 of 1193, across 6 files
- Merge hook into existing settingsin 21 of 1193, across 3 files
- Ask if installation is global or projectin 20 of 1193, across 2 files
- Copy the hook script to target locationin 20 of 1193, across 2 files
Said here and by no other author read
- Read the provided task file
- Extract the task file frontmatter
- Load the specified sub-skill
- Execute the extracted task
- Keep the result under eighty lines
- Note any assumptions in the result
Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.