skills.rest
pdw96
pdw96COMMUNITY
Explore all skills by this author →
GitHub

authority-line

Defines which repository actions agents perform autonomously and which require human approval.

Updated Aug 31, 2026
Search Tags:#code review#authorization#pull request#github#workflow rules#agent governance#session orchestration
One-click install
npx skills add https://github.com/pdw96/ai-production-risk-early-warning --skill authority-line-pdw96
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: authority-line
Source: https://github.com/pdw96/ai-production-risk-early-warning/tree/main/.claude/skills/authority-line
Command: npx skills add https://github.com/pdw96/ai-production-risk-early-warning --skill authority-line-pdw96

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? When AI agents work on a repository, it is unclear which actions they may take without asking and which must be left to a human, leading to risky operations like merging pull requests or requesting reviews without authorization. This Skill codifies the authority boundary so agents act consistently across sessions. ## Core Features & Use Cases - Autonomous action list: Spells out what agents may do without asking, such as spawning child sessions with task briefs, sending instructions to running sessions, and subscribing to PR events. - Human-reserved actions: Enumerates operations reserved for the repository owner, including merging, review requests, draft/ready transitions, closing PRs, resolving review threads, and stage-kickoff decisions. - Named forbidden endpoints: Explicitly bans the enable_pr_auto_merge tool and four CCR API paths (resolve/unresolve, auto_merge, ready_for_review, convert_to_draft) even when technically reachable. - Use Case: An agent finishes a task and considers marking its PR ready for review; consulting this Skill tells it that ready-for-review is a human action and it must stop instead. ## Quick Start Consult the authority-line skill before merging a pull request, requesting a review, or starting the next work stage to confirm whether that action belongs to the agent or the human owner.

Frequently Asked Questions about authority-line

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
Which GitHub actions can an AI agent perform without asking?▼

The agent may spawn child sessions with task briefs and action limits, send instructions to running sessions via scheduling, subscribe to and unsubscribe from PR events, and create or reschedule check-ins. Everything else involving merges or reviews is reserved for the human owner.

Can an AI agent merge a pull request if the button is technically clickable?▼

No. Merging is a discipline rule, not a technical lock. Measurements showed the same merge call was blocked one day and succeeded another, so the rule is that the agent chooses not to merge regardless of whether the API allows it.

Why is enable_pr_auto_merge forbidden for agents?▼

Calling enable_pr_auto_merge is treated as de facto agent merging, since it changes when the merge executes without a human pressing the button. The same applies to four CCR API paths covering thread resolution, auto-merge, ready_for_review, and convert_to_draft.

When should an agent notify the human owner during a task?▼

Only in three cases: when work is finished, when it is blocked, or when an assumption turns out to be wrong. Progress reports do not qualify because every wake-up consumes turns from a shared five-hour limit across all lanes.

What is the context threshold for switching agent sessions?▼

When a session's context usage exceeds 40 percent, no new work is assigned to it. The current task is finished, and the next piece of work goes to a fresh session, since the owner considers 70 percent already too much.

Should an agent start the next project stage on its own?▼

No. An open issue means the work was decided, not that it is the agent's turn to start it. The owner decides what to pick up and when, even after all milestones in the current stage are closed.

Related Skills

Explore More Software Engineering Skills→
j-mckerracherj-mckerracher

scope-and-security

Enforce file access and action boundaries for AI agent workflows.

Community
Advanced
compozycompozy
2.7k

cy-orchestrate-tasks

Orchestrates spec tasks by spawning bounded worker sessions and verifying on-disk completion.

Official
Advanced
automateyournetworkautomateyournetwork
650

globalping-external-checks

Measure public endpoints from thousands of global probes using ping, traceroute, DNS, MTR, and HTTP.

Community
Intermediate
groeimetaigroeimetai
77

acl-security

Create and configure record and field-level ACLs in ServiceNow.

Community
Advanced