Skip to content
Agentmart
Submit a tool

human-itl

Walk a human through the manual-action [HITL] sub-tasks of a parent (PRD) issue — the steps a coding agent physically cannot perform — recording each as an…

Skill by ZunoSmartLabs

IndexedMIT★ 2
Write a review

Install

Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.

npx agentmarket install skill/zunosmartlabs-human-itl

Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.

Get the weekly diff

The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.

We use your email only for this newsletter. Privacy

About

Walk a human through the manual-action [HITL] sub-tasks of a parent (PRD) issue — the steps a coding agent physically cannot perform — recording each as an audit trail and marking them done so the dependent [AFK] slices unblock, then hand back to /tdd-parallel. Use when /tdd-parallel skipped HITL slices, when the user wants to clear manual steps before a fanout, or when the user invokes /human-itl directly.

Use cases: testing, security, agent-building

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.

Tools
No tools found
Secrets
None asked for
Install scripts
None found

Network hosts it talks to 1

  • codeload.github.com · found in install command

Files it touches

  • ~/.claude/skills/human-itl

What's inside

Skill name
human-itl
Description
Walk a human through the manual-action [HITL] sub-tasks of a parent (PRD) issue — the steps a coding agent physically cannot perform — recording each as an audit trail and marking them done so the dependent [AFK] slices unblock, then hand back to /tdd-parallel. Use when /tdd-parallel skipped HITL slices, when the user wants to clear manual steps before a fanout, or when the user invokes /human-itl directly.
License
MIT
Instructions
107 lines in SKILL.md · view on GitHub

Files (1)

  • SKILL.md

SKILL.md (first lines)

# Human in the loop

`/tdd-parallel` fans out only the **`[AFK]`** slices of a PRD — the ones an
agent can run unattended. The **`[HITL]`** slices it skips need a human's
hands or authority. This skill is the serial, human-present counterpart:
it walks you through each manual-action slice, records what you did, and
marks it done so the `[AFK]` slices that were `Blocked by` it unblock.

It does **not** write code or run red-green-refactor — that's `/tdd`. A
`[HITL]` slice is a manual *action*, not a unit of TDD work.

## What counts as HITL

A legitimate `[HITL]` slice requires a manual action a coding agent
**physically cannot perform**:

- Clicking through a third-party console (DNS registrar, cloud dashboard).
- Rotating or issuing a real credential / secret / certificate.
- Obtaining external sign-off (legal, security, a stakeholder).
- Running a one-off production migration or cut-over by hand.

It is **not** an architectural decision or a design review. Those must be
resolved upstream via `/grill-with-docs` and recorded as ADRs *before*
issues are cut, so slices fall out maximally AFK — see `/to-issues`.

**Hard refuse decision slices.** If a slice is really a decision in
disguise (`Decide auth provider`, `Pick the cache strategy`), do **not**
walk the user through it. Flag it as a process leak and stop on that
slice: tell the user to resolve it with `/grill-with-docs` + an ADR, then
relabel it and its dependents `[AFK]` and re-run the fanout. Continue with
the remaining genuine `[HITL]` slices.

If a "HITL" slice turns out to be implementable code, it's mislabeled —
say so and tell the user it belongs in the AFK fanout (`[AFK]`).

## Usage

```
/zsl:human-itl <parent-issue>   # clear all HITL slices of a PRD
/zsl:human-itl <hitl-issue>     # clear one HITL slice
/zsl:human-itl                  # no arg → picker of open HITL slices
```

## Process

### 1. Pre-flight

- `docs/agents/issue-tracker.md` exists (run `/setup-zsl-superpowers` if
  not). `docs/agents/triage-labels.md` exists.
- Resolve the input. A parent issue → its sub-issues. A single issue →
  just that one. No argument → fetch open `[HITL]` issues carrying the
  configured `ready-for-agent` label and present a numbered picker.
- Refuse a container issue (one with open sub-issues that is itself
  titled `[HITL]`) the same way the other skills do.

### 2. Discover the slices

Using `docs/agents/issue-tracker.md` conventions, fetch the parent's
sub-issues. Keep only those that are:

- Open
- Title starts with `[HITL]`
- Carry the configured `ready-for-agent` label
- Not a container

Parse each slice's `## Blocked by` section. A `[HITL]` slice may itself be
blocked by an `[AFK]` slice that a prior fanout already merged — process
slices in dependency order, skipping any whose blockers aren't satisfied
yet (report them as "still blocked").

### 3. Walk the slices

Show the user the buckets first — **To clear** (numbered), **Still
blocked** (issue + unmet blockers), **Refused — decision slice** (issue +
why) — and confirm before starting.

Then, for each slice in dependency order:

1. Present the slice: title, the `## What to build` action, and the
   `## Acceptance criteria` ("Done when") checklist.
2. **Pause.** The user performs the manual action out of band.
3. Ask the user for a one-line confirmation note (what they did; any
   id/link/ticket as evidence). Post it as a comment on the issue per
   `docs/agents/issue-tracker.md` — this is the audit trail.
4. If the action embodies a decision worth keeping (a chosen provider, a
   negotiated limit), write or update an ADR — lazily, like
   `/grill-with-docs`. Don't batch; capture it now.
5. Mark the issue done per the tracker convention (close it / `git mv`
   into `issues/_done/` for local-markdown). This is what unblocks the
   dependent `[AFK]` slices.

### 4. Finish

Print a summary: slices cleared (issue + note), slices still blocked,
slices refused as decision leaks. Then **stop** with the explicit next
step — do not chain:

> Cleared N HITL slices. Run `/tdd-parallel <parent>` to fan out the
> now-unblocked AFK work.

## Not in scope

- Writing code or running red-green-refactor — that's `/tdd`.
- Resolving decisions — that's `/grill-with-docs` + an ADR, upstream.
- Invoking `/tdd-parallel` for the user — this skill ends with the hint.

From ZunoSmartLabs/zsl-superpowers, MIT.

More from ZunoSmartLabs/zsl-superpowers

Signals

Stars
2
Forks
0
Open issues
0
Last commit (this folder)
Jun 1, 2026 (5 months ago)
Last release
Sep 18, 2026 (1 month ago)

From GitHub (fetched Oct 24, 2026).

Details

License
MIT
First seen
Oct 24, 2026
Last checked
Oct 24, 2026

More from ZunoSmartLabs

Cite this page

human-itl skill. Agent Mart. https://agentmart.dev/skills/zunosmartlabs-human-itl (last updated 2026-10-24)