openloops-needs-human

Files agent blockers as labeled GitHub issues carrying verifiable shell predicates.

Updated Aug 25, 2026
One-click install
npx skills add https://github.com/thorwhalen/openloops --skill openloops-needs-human-thorwhalen
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: openloops-needs-human
Source: https://github.com/thorwhalen/openloops/tree/main/openloops/data/skills/openloops-needs-human
Command: npx skills add https://github.com/thorwhalen/openloops --skill openloops-needs-human-thorwhalen

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires openloops.

What problem does it solve? When an AI agent gets blocked on something only a human can do — setting a secret, running a privileged command, granting access, making a decision — that obligation normally dies in a chat transcript nobody re-reads. This Skill turns the blocker into a durable, queryable manual-task GitHub issue that outlives the session and can be re-checked automatically. ## Core Features & Use Cases - Structured blocker filing: Creates a manual-task issue in the correct repository with a required Ask, Verify, Evidence, and Blocks field set, plus duplicate detection and a fallback-repo rule when you lack labeling permissions. - Verifiable completion predicates: Every issue carries a shell command whose exit status reports whether the ask is done, so ol owed can re-check obligations discharged out of band instead of leaving issues open forever. - Mandatory egress scrubbing: A built-in Python check rewrites home paths, refuses to file on credential-shaped text, unfilled placeholders, or unrewritable absolute paths, and supports reopening recorded decisions (ADRs) with a five-field template. - Use Case: An agent cannot publish a package because PYPI_PASSWORD is missing. It files an issue titled "Set PYPI_PASSWORD on the release repo" with a gh secret list verify predicate, scrubs the draft, and links the issue in its closing message. ## Quick Start Tell your agent: file a needs-human issue for the missing CI secret on this repo so I can see it later with ol owed.

Frequently Asked Questions about openloops-needs-human

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

FAQPage Schema
How do I file a GitHub issue when an AI agent is blocked on a human?▼

Run the skill's single Bash block: it writes the title and body, scrubs paths and credentials with openloops.egress, then calls gh issue create with the manual-task label. The body must include Ask, Verify, Evidence, and Blocks fields.

What is a verify predicate in a manual-task issue?▼

A verify predicate is a shell command in the **Verify:** field whose exit status is 0 once the ask is done, such as checking gh secret list for a required secret. It lets ol owed re-check obligations discharged out of band instead of leaving issues open forever.

When should an agent not file a manual-task issue?▼

Do not file when the agent could do the task itself, when the ask is a preference where any reasonable choice works, or for routine review requests. Try the action once first; only permission, credential, or access failures justify filing.

What happens if the draft contains a credential or secret?▼

The egress check refuses to file and exits non-zero. The agent must ask the human for safe wording rather than redacting, and must never quote the matched secret text in the issue, a comment, or a message.

Where is the issue filed if the agent lacks write permission on the affected repo?▼

It is filed in a human-nominated fallback repo with an Affected repo line naming the real one, and the title starts with the affected repo name. Nothing is ever posted into a client or another team's tracker without approval.