Index of every page on this site: llms.txt
All skills
Run it yourself

Track every action only a human can take — a login, a session approval, a phone-side action, a platform identity check, or a decision only the human can make — so it never ages silently. Each gate is logged once in a gate log file, re-surfaced on a schedule, and closed only on verification. Use whenever your work hits a step that only the human can do, at the start of any session that touches blocked work, and when deciding whether an old gate stays open or gets dropped.

  • 4 files
  • 12.1 KB
  • Updated 5 days ago
  • GitHub
Use this Skill: https://skilld.dev/gh/jokerhtml/needcraft-skills/human-gates

Nothing lands on disk. Nothing to clean up.

Fork this Skill

Edit a local copy. It keeps the author and licence.

SKILL.md

≈123 tokens for metadata: the name and description. ≈1.9k when used: this file. ≈815 more on demand in 2 files.

Description uses 6.1% of example budget

Before choosing a Skill, your Agent reads its name and description. All available Skills share that space.

  • A shorter description leaves more room for other Skills. This entry exceeds our 1% size suggestion.

In our Claude Code example, all Skill names and descriptions share 8,000 characters. This Skill uses ≈489 characters, or 6.1%.

The 1% threshold is a size suggestion. Longer descriptions can still fit.

Your model, settings, and other Skills decide how much text your Agent can read.

Example settings and source

The example uses a 200k-token context and default Claude Code settings. The count includes the name, description, separators, and when_to_use when present. Codex also counts local file paths.

Skit's source and limits: Codex 0.160.1, Claude Code 2.1.292.

Human Gates

What this skill is

A human-in-the-loop tracker. Some steps can only be done by the human who owns the account or the decision: signing in, approving a session, tapping something on their phone, passing an identity check, or making a call only they can make. This skill makes sure those steps get asked once, logged once, re-surfaced on a schedule, and closed only when verified — so a gate asked on Monday is not silently forgotten by Friday.

Who it is for: anyone (human or agent) doing work on someone else's accounts, where some steps need the account owner. Also useful for a solo operator tracking their own blocked steps across sessions.

Prerequisites

  • One gate log file, chosen by you. A plain text or Markdown file in a place every session and every scheduled run can read — for example gates.md in your project notes, or a section of your main notes file. The whole skill runs on this one file. Pick it once, write its path at the top of your notes, and never move it.
  • A way to remind yourself: a calendar reminder, a scheduled check, or a session-start habit (see "Setup" below). The skill does not depend on any particular reminder tool.

When to use

  • The moment your work hits a step only the human can do.
  • At the start of any session or run that touches blocked work — read the log first.
  • When a gate has waited 3, 7, or 14 days (see "Re-surface the gate").
  • When the blocked work can proceed partly — stage everything around the gate.

Setup

  1. Create the gate log file. Put this header at the top:
# Human gates

One line per gate. Never rewrite history, never merge two gates into one line.
Columns: opened date | gate (one line) | exact steps for the human | asked dates | status (OPEN/CLOSED/DROPPED) | resolution + how verified
  1. Make reading the log a habit: read it at the start of every session and every scheduled run that touches the blocked area, and set a recurring reminder (daily or weekly, your choice) to review open gates. The reminder tool is yours — a calendar event, a cron job, a scheduled agent run, or a sticky note. The skill needs the review to happen; it does not care what triggers it.
  2. Tell the human, once, where the log lives and that open gates will be re-surfaced — so the reminders are expected, not a surprise.

The blocker protocol (summary)

Before opening a gate, confirm the blocker is real. The rule:

  1. Try at least 3 genuine attempts before declaring a blocker — across sessions when the failure looks transient. A flaky session is not a gate. A genuine attempt means waiting for the page or tool to fully load, reloading once, and checking the actual state (not assuming from an error message).
  2. Then escalate to the human with the exact ask, exact steps, and a timebox. The ask goes in chat first; the log entry is the record, never a replacement for the ask.
  3. Never work around the gate silently. Do not find a different route that bypasses a login, an approval, or an identity check the human is supposed to do. If a workaround exists and is legitimate, say so explicitly and let the human decide.
  4. Stage everything around the gate. While it waits, do all the work that does not need it, so resolving the gate later needs no rework. Say plainly what is staged, what is blocked, and what resolving the gate unblocks.

Open a gate

  • Log the gate in the gate log file the moment you ask the human, before you move on. Record the date, the gate in one line, the exact steps for them, and the status OPEN.
  • Report the ask in chat first, with exact steps and a timebox. The log is the queue; it never replaces the ask.
  • Keep one line per gate and append every asked date. Never rewrite history or merge two different gates into one line.

Example gate entry

- 2026-10-06 | Social profile link needs updating from the phone app (app-only setting) | 1. Open the app on your phone. 2. Go to Edit profile. 3. Set the website field to the new page URL. 4. Tell me when it is done so I can verify it live. | Asked: 2026-10-06, 2026-10-09 | OPEN

Re-surface the gate

  • Read the log at the start of every session and every run that touches the blocked area. Check which gates are still open and how long they have waited.
  • After 3 days open: re-surface the gate with one chat line carrying the exact ask again. Mention the days waited so the human sees the age.
  • After 7 days open: collect every stale gate into one short list, newest first, and post it at most once per day. Stop attaching gate asks to unrelated conversations.
  • After 14 days open: bring the gate to chat for a keep-or-drop decision. A dropped gate leaves the log with a one-line reason and its asked history intact.

The 3 / 7 / 14-day cadence is a tunable default, not a fixed rule. It assumes a human who checks chat roughly daily and work that can wait a few days. Tighten it (1 / 3 / 7) if the gate blocks a launch or a deadline; loosen it (7 / 14 / 30) if the human checks in weekly and nothing is time-critical. Whatever cadence you choose, write it in the log header and stick to it — the point is that re-surfacing is scheduled, not improvised.

Close the gate

  • Close a gate when the human confirms the action, when the action verifies live, or when the need disappears. Record the resolution date and how you verified it.
  • Never close on assumption or because the run moved on. An unverified gate stays OPEN.
  • A dropped gate (after the 14-day keep-or-drop decision) leaves the log with a one-line reason and its asked history intact.

Example closed entry

- 2026-10-06 | Social profile link needs updating from the phone app (app-only setting) | 1. Open the app on your phone. 2. Go to Edit profile. 3. Set the website field to the new page URL. 4. Tell me when it is done so I can verify it live. | Asked: 2026-10-06, 2026-10-09 | CLOSED 2026-10-10 — human confirmed; verified live on the profile page

Copy-paste template

Use this to open a gate in chat and in the log:

GATE — [one-line gate]
Needed from you:
1. [exact step]
2. [exact step]
3. [exact step]
Time needed: [minutes]
Blocks: [what is staged and waiting on this]
If this waits: [what happens if it stays open]
Asked: [date]

Failure modes

  • A gate asked once and never re-surfaced. Mitigate: the 3 / 7 / 14-day re-surface schedule, and the session-start log read.
  • A gate closed on assumption. Mitigate: close only on confirmation or live verification, with the verification method recorded.
  • Two gates merged into one line. Mitigate: one line per gate; a new ask is a new line, even if it looks similar.
  • A workaround applied silently. Mitigate: the blocker protocol — no bypass without the human's explicit decision.
  • The log lives in two places. Mitigate: one log file, chosen once, named in your notes. If you ever move it, leave a pointer at the old location.

Known limits

  • This skill tracks gates; it does not resolve them. A gate closes when the human acts or the need disappears — no tracking system changes that.
  • The cadence defaults are a starting point. A human who checks chat weekly needs a different schedule than one who checks hourly; tune it and write the tuned version in the log header.
  • If the human never responds, the skill's answer is the 14-day keep-or-drop decision — not louder or more frequent asking.

Source: SKILL.md on GitHub

No rule matched.

skilld matched fixed text patterns in SKILL.md and file names. Patterns miss obfuscated code.

skilld run checks every file with the same patterns. It asks for approval before it loads a Skill with a behavior marked Needs approval.

No third-party reports yet.

Signed by skilld at 90f40b1. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub 4 days ago.

Activeupdated 5 days ago
Other metadata
metadata
{
  "version": "1.0.0",
  "date": "2026-11-16",
  "author": "Muhammad Saad Ranjha / NeedCraft"
}

README badge

README badge for jokerhtml/needcraft-skills/human-gates