Manages authorization scopes delegating autonomous entity work to AI agents.
npx skills add https://github.com/23min/aiwf --skill aiwf-authorize-23min
Please help me install this Agent Skill. Skill: aiwf-authorize Source: https://github.com/23min/aiwf/tree/main/internal/skills/embedded/aiwf-authorize Command: npx skills add https://github.com/23min/aiwf --skill aiwf-authorize-23min
aiwf authorize <id> --to <agent>, then pause, resume, or end them with recorded reasons, each producing a single commit with structured trailers. - Provenance enforcement: Agent-mode commits are decorated with aiwf-on-behalf-of: and aiwf-authorized-by: trailers; missing or ended scopes surface as provenance-no-active-scope and related check findings. - Automatic scope termination: Promoting an entity to a terminal state or cancelling it auto-ends all non-ended scopes on that entity. - Use Case: A user says "implement E-0042 autonomously on branch feat/cache". The assistant runs aiwf authorize E-0042 --to ai/claude --branch feat/cache, then performs the work; every commit it makes is provably tied back to that human grant. ## Quick Start Ask the assistant to authorize autonomous work by saying something like "authorize ai/claude to work on E-0042 autonomously" and it will open the scope with the aiwf authorize command.High-intent search queries and answers about installing and using this skill.
Run `aiwf authorize <id> --to <agent>` where the id is the entity and the agent is an operator id like ai/claude. This opens an active scope, and subsequent agent commits carry an aiwf-authorized-by trailer referencing it.
Tool mode is human-in-the-loop: verbs run with --actor ai/<id> --principal human/<id> and need no scope. Agent mode is autonomous: it requires an open authorization scope, and commits gain aiwf-on-behalf-of and aiwf-authorized-by trailers.
No. The authorize verb requires a human actor and refuses non-human actors with a usage error. Only humans can grant authority, per the kernel's rule that force is sovereign to humans.
This finding means the agent ran a verb without a matching active authorization scope. Open one with `aiwf authorize <id> --to <agent>`, resume a paused scope, or run the verb directly as the human principal instead.
Promoting the scope-entity to a terminal state or cancelling it auto-ends every non-ended scope by writing an aiwf-scope-ends trailer. Ended scopes cannot be resurrected; issue a fresh authorize command to restart work.
LSDIPPOLLCTransform feature scope into sequenced Linear issues with hierarchy and dependencies.
BlackBeltTechnologyAutomate local release preparation by updating CHANGELOG.md, bumping package.json, and pushing version tags.
srdjanQuery git commit trailers to surface decision context and rationale.
dotlab-hqConfigure human-in-the-loop approval for LangChain tool calls with per-tool policies.