askit-capability-advisor
Purpose
Tell a maintainer what their target agent supports before they build, so the plan fits the agent rather than failing at release. Two modes: advise takes the target agents (claude, codex, or both), reports which component types are plugin-distributable on each, flags the Claude-only types and the cross-agent asymmetries, and recommends a tier path; check takes a specific component or a whole plugin and reports which declared targets can actually run it, so a Claude-only component shipped under a Codex target is caught before release. The full matrix is in foundation/synthesis/capability-matrix.md.
When to use
When choosing agent-targets, checking whether a component is portable across agents, or deciding which conformance tier to aim for.
advise mode
- Read the target agents (
claude, codex, or both).
- For each, report the plugin-distributable component types and the asymmetries: subagents are Claude-only for plugin distribution (Codex ingests them only via user/project
config.toml, Standard sec 3.3); output styles are Claude-only and the statusline differs (sec 2.3); a command's Codex form is its backing skill, explicitly invocable, not a separate artifact (sec 3.2); hooks are a Codex subset of Claude's event set (sec 3.5); skills, MCP, and AGENTS.md are portable on both (sec 3.1, 3.9, 3.10).
- Recommend a tier path: Bronze (skills + references + AGENTS.md + MCP), Silver (adds subagents, commands, workflows, chain contracts, plugin packaging, the prefix, native manifests), Gold (adds hooks, output styles, self-hosting CI).
check mode
- Read the component (or every component in the plugin) and the plugin's declared
agent-targets.
- For each declared target, report whether the type is plugin-distributable there. Flag a Claude-only type (subagent, output style, statusline) carried under a
codex target, since it cannot ship to Codex from a plugin.
- Point each flag at the fix: set a per-component
agent-targets: [claude] override (sec 3.7), or move the component into a Claude-only plugin.
Scope
Advisory only: it reports capability and recommends a tier; it does not author components (the askit-build-* skills do) or grade conformance (askit-evaluate does). The matrix tracks the agent versions the Standard pins per component (the load-bearing Codex subagent constraint is pinned to Codex CLI v0.135 in sec 3.3); when an agent adds a capability the matrix and this skill update together, so the advice never drifts ahead of what the agents actually support.
1---2name: askit-capability-advisor3description: Reports which component types a target agent can run and recommends a conformance tier before a plugin is built, mapping Claude Code and Codex capabilities to the Advanced Skill Library Standard's component types. Use when choosing agent-targets, checking whether a component is portable across agents, or deciding which tier to aim for.4---56# askit-capability-advisor78## Purpose9Tell a maintainer what their target agent supports before they build, so the plan fits the agent rather than failing at release. Two modes: `advise` takes the target agents (`claude`, `codex`, or both), reports which component types are plugin-distributable on each, flags the Claude-only types and the cross-agent asymmetries, and recommends a tier path; `check` takes a specific component or a whole plugin and reports which declared targets can actually run it, so a Claude-only component shipped under a Codex target is caught before release. The full matrix is in [`foundation/synthesis/capability-matrix.md`](../../foundation/synthesis/capability-matrix.md).1011## When to use12When choosing `agent-targets`, checking whether a component is portable across agents, or deciding which conformance tier to aim for.1314## advise mode151. Read the target agents (`claude`, `codex`, or both).162. For each, report the plugin-distributable component types and the asymmetries: subagents are Claude-only for plugin distribution (Codex ingests them only via user/project `config.toml`, Standard sec 3.3); output styles are Claude-only and the statusline differs (sec 2.3); a command's Codex form is its backing skill, explicitly invocable, not a separate artifact (sec 3.2); hooks are a Codex subset of Claude's event set (sec 3.5); skills, MCP, and AGENTS.md are portable on both (sec 3.1, 3.9, 3.10).173. Recommend a tier path: Bronze (skills + references + AGENTS.md + MCP), Silver (adds subagents, commands, workflows, chain contracts, plugin packaging, the prefix, native manifests), Gold (adds hooks, output styles, self-hosting CI).1819## check mode201. Read the component (or every component in the plugin) and the plugin's declared `agent-targets`.212. For each declared target, report whether the type is plugin-distributable there. Flag a Claude-only type (subagent, output style, statusline) carried under a `codex` target, since it cannot ship to Codex from a plugin.223. Point each flag at the fix: set a per-component `agent-targets: [claude]` override (sec 3.7), or move the component into a Claude-only plugin.2324## Scope25Advisory only: it reports capability and recommends a tier; it does not author components (the `askit-build-*` skills do) or grade conformance (`askit-evaluate` does). The matrix tracks the agent versions the Standard pins per component (the load-bearing Codex subagent constraint is pinned to Codex CLI v0.135 in sec 3.3); when an agent adds a capability the matrix and this skill update together, so the advice never drifts ahead of what the agents actually support.
product-on-purpose