From ai-assistant-ops
Use when adapting skills for AI Assistant Harness Adaptation, using an assistant harness action matrix, or making explicitly named target skills harness-agnostic.
How this skill is triggered — by the user, by Claude, or both
Slash command
/ai-assistant-ops:adapt-skill-for-ai-harnessThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
This skill adapts an explicitly named target skill from one AI assistant harness vocabulary to another.
This skill adapts an explicitly named target skill from one AI assistant harness vocabulary to another.
The assistant harness action matrix is the reusable source of truth, not per-harness explanatory notes for this skill.
matrix = load_json(metadata["ai-assistant-harness-adaptation.action-matrix"])
entry = matrix["actions"][action_key][assistant_key]
codex_agent = matrix["actions"]["CreateAgent"]["Codex"]
Use scripts/lookup_harness_action.py --action CreateAgent --assistant Codex for scriptable lookup.
Adapt only an explicitly named target skill or target skill path. Do not migrate every skill in a repository unless the user explicitly asks for that broader migration.
The matrix maps shared-purpose workflows, skills, tools, and commands that AI harnesses name differently. It should capture names, aliases, surfaces, discovery paths, and short translation nuances, not explain a harness to itself.
Adapted target skills may still receive references/ai-assistant-harnesses/claude-code.md and references/ai-assistant-harnesses/codex.md when that keeps target-specific wording out of shared SKILL.md.
Use Claude Code format, wording, and tool names as the baseline in shared target SKILL.md content.
For target skills, add flat metadata links, one per supported harness:
metadata:
ai-assistant-harness-adaptation.claude-code: references/ai-assistant-harnesses/claude-code.md
ai-assistant-harness-adaptation.codex: references/ai-assistant-harnesses/codex.md
Metadata values must be file links only. Do not store tool mappings, prose policy, compatibility notes, or lists of all harness instructions there.
Store target harness adaptation files under references/ai-assistant-harnesses/<harness-id>.md. Keep them narrow and include only actions the target skill needs.
The adapted target SKILL.md must tell assistants to identify the active harness, load exactly one matching metadata-linked harness reference when harness-specific adaptation is needed, and skip non-matching harness files.
Do not create a shared cross-harness instruction table or a full list of all assistant harness instructions in the shared target SKILL.md.
Do not create references/ai-assistant-harnesses/ inside this skill. That directory is an output pattern for adapted target skills.
references/harness-action-matrix.json when harness-specific mappings are needed.SKILL.md, files under references/, files under agents/, examples, assets directly loaded or described to the model, scripts referenced by the skill, and similar explicitly loaded files.CreateAgent, AskUser, TrackTasks, EditFile, or SlashCommand.matrix["actions"][action_key][assistant_key]; for example, use matrix["actions"]["CreateAgent"]["Codex"].SKILL.md.references/ai-assistant-harnesses/codex.md.references/ai-assistant-harnesses/claude-code.md.references/ai-assistant-harnesses/ format and remove duplicated old adaptation information.SKILL.md.Keep action keys stable and TitleCase. Keep assistant keys stable and product-oriented, currently ClaudeCode and Codex.
Each action must have a kind value: workflow, skill, tool, or command.
Prefer compact fields:
surface: where the concept appears, such as tool, skill, slash-command, or workflow.terms: current names the harness uses.aliases: older or alternate names.discovery: how to find deferred or optional capabilities.nuances: short compatibility notes that matter during translation.Preserve lookup_order: ["action", "assistant"] so scripts can keep using matrix["actions"][action_key][assistant_key].
Do not batch-migrate unrelated skills.
Do not rewrite README, tests, docs, or catalog text while adapting a target skill unless the user-facing catalog or verification changes require it.
Do not load other target harness references while acting as a specific harness.
Do not leave both old per-assistant instructions and new metadata-linked harness files active for the same behavior.
Do not replace the Claude Code-baseline shared workflow with Codex wording. Put Codex wording in the target codex.md.
For each adapted target, check that:
SKILL.md has only flat metadata.ai-assistant-harness-adaptation.* links..md and resolves under references/ai-assistant-harnesses/.For this skill itself, check that:
metadata.ai-assistant-harness-adaptation.action-matrix resolves to references/harness-action-matrix.json.matrix["actions"][action_key][assistant_key] lookup works for at least CreateAgent and Codex.references/ai-assistant-harnesses/claude-code.md or references/ai-assistant-harnesses/codex.md.Run the repository's relevant static tests, eval checks, JSON validation, and Markdown whitespace checks before finishing.
Creates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.
npx claudepluginhub alex-kopylov/zweihander --plugin ai-assistant-ops