Skip to content

wandb/senpai / plugins/senpai/skills/check-human-issues

check-human-issues

Verified

Open GitHub Issues for human input and respond to the researcher team. Use this skill whenever you need to handle a human_issue event, respond to human issues, ask humans a question, or check team communications. Also triggers for: "any human messages?", "check issues", "respond to humans".

FileKindSize
SKILL.mdskill md3.6 KB
---
# SPDX-FileCopyrightText: 2026 CoreWeave, Inc.
# SPDX-License-Identifier: Apache-2.0
# SPDX-PackageName: senpai

name: check-human-issues
description: >
  Open GitHub Issues for human input and respond to the researcher team.
  Use this skill whenever you need to handle a human_issue event, respond to
  human issues, ask humans a question, or check team communications. Also triggers
  for: "any human messages?", "check issues", "respond to humans".
argument-hint: "<name> <ADVISOR|STUDENT>"
---

# check-human-issues

Check GitHub Issues tagged `human` for messages from the research team, and respond to any that need a reply.

## Arguments

- **$0** — The configured advisor branch for `ADVISOR`, or student name for
  `STUDENT`
- **$1** — Either `ADVISOR` or `STUDENT`

## How it works

Human researchers communicate with agents through GitHub Issues. Issues are
tagged with `human` plus `team` for a broadcast, the configured advisor branch
for an advisor, or `student:<student-name>` for a student. Your job is to check
messages routed to your exact role, respond to new ones, and skip ones you've
already handled.

## Open an issue for human input

Use `create_human_issue` when you need a research decision or help with a blocker.
Provide a stable `issue_id`, a concise `title`, and the complete `body` without a
role prefix. Reuse the same ID, title, and body if the call needs a retry. Changed
content requires a new ID.

The runtime adds `human` and your configured audience label. It mentions the
GitHub credential owner in the initial issue body when GitHub identifies that
owner as a user account. This includes service accounts registered as ordinary
users. Bot identities, tokens without a user identity, and failed lookups skip
the mention without blocking issue creation. Do not add notification mentions
yourself. Later replies do not repeat these automatic mentions.

```json
{
  "issue_id": "confirm-next-seed-budget",
  "title": "Confirm the budget for another seed",
  "body": "The first two seeds disagree. Can we use the remaining budget for a third seed?"
}
```

## Steps

1. **Read the current `human_issue` event.** The controller supplies the issue
   identity and the exact human message ID that triggered the wake. It polls
   GitHub for issues addressed to you or the whole team.

2. **Decide whether to respond:**
   - If you haven't commented on this issue yet → respond.
   - If you have commented, check if the human posted a new comment *after* your last response. If so → respond to the new message. If not → skip, you're waiting for the human.
   - Record the exact numeric `id` of the issue body or human comment you are
     answering. Never substitute the issue number for a comment ID.

3. **Respond** through `respond_to_human_issue` with the issue number, the exact
   `human_message_id`, and the response text without a role prefix. This verified,
   idempotent operation refuses closed issues, pull requests, missing `human`
   labels, stale message IDs, Senpai protocol messages from the agent identity, and
   issues not addressed to this configured advisor branch or student.

```json
{
  "issue_number": 123,
  "human_message_id": 987654,
  "response": "<your response>"
}
```

Never mutate the issue through `gh` or `curl`.

4. **Never close human issues.** Only the human does that.

## Return format

When you're done, record a structured summary of the issues you checked and
responded to in the current conversation:

### New research directives from the human researcher team

If there are research directives in the issues, include them in detail so they
remain available for subsequent planning in this conversation.
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 472kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 322kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 322kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 322kUpdated Claude CodeCodexVerifiedInstall