Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add skills/haiggoh/run-to-completion/ungate-queuenpx skills add haiggoh/run-to-completion --skill ungate-queuegit clone --depth 1 https://github.com/haiggoh/run-to-completionWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/haiggoh/run-to-completion/ungate-queue)<a href="https://agentmods.dev/skills/haiggoh/run-to-completion/ungate-queue"><img src="https://agentmods.dev/badge/skills/haiggoh/run-to-completion/ungate-queue.svg" alt="Measured on agentmods" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00132 | $0.02215 |
| Opus 5 | $0.00066 | $0.01107 |
| Sonnet 5 | $0.00026 | $0.00443 |
| Haiku 4.5 | $0.00013 | $0.00221 |
Grade A, and why
ungate-queue scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured yesterday.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 101 lines — stays where its author put it; the contents beside it link to each section on GitHub.
ungate-queue — remove the gate, then stop
An unattended run produces two piles. The autonomous pile gets done without the user. This skill is the other half: the gated pile, walked WITH the user. The user's presence is the whole premise, so nothing here is deferred or guessed.
This pass spends user attention, not compute. That is what makes it the right pass to run when compute is scarce or expensive — but only while it stays cheap. The moment it starts doing the work it unblocked, it costs both, and the property that made it worth running is gone.
The boundary: record and return
Your output is a changed gate state, not a changed repository. For each item: ask, record the answer on the item, retier it, move to the next one. Then hand the released pile to a later unattended run.
Concretely, per item you are done when:
- The answer is written onto the item itself, in enough detail that a fresh session could act on it without you in the room.
- The item's verdict is updated through the queue's own interface — no longer gated, or still gated with a smaller reason.
- You have moved on.
You are NOT done-and-continuing. Do not open the files. Do not start the fix "while we are here". If an item is one line of work and the temptation is strong, that is exactly the case to resist: releasing thirty gates in one attended sitting is worth far more than releasing four and fixing one.
Only the FIRST gate is yours. An item often has a second blocker behind the first. Remove the one in front, record what is now known about the next, and let the item come back around. Chasing a chain to its end turns one question into a project.
Rank the gated pile by how cheaply it can be released
Gated is not one flat bucket. Items differ enormously in what an answer buys, so present the cheapest-to-release first. Ask in this order:
- G1 — one answer, gone forever. A single answered question removes the gate permanently. Highest value per unit of attention. These are what this pass exists for; ask them first, always.
- G2 — an installation something already does. The gate is a setup step an existing script or documented command already performs, so it costs little or nothing to clear. Still cheap, still good.
- G3 — a custom installation. Clearing it needs real work: trying approaches, testing, iterating. Genuinely harder, so it must not crowd out G1 and G2. Surface it after them.
- G4 — not an ungating candidate at all. The gate is constant multi-step user interaction while the task runs. There is no answer that releases it into autonomy; the task simply has to be done attended. Keep these at the back and raise them only if the user explicitly asks. Naming one as attended work is a complete result.
- ENV — a recurring precondition, not a failure. An external condition that will be true sometimes and false at other times: an application must be running, a particular network must be joined, a device must be plugged in. These can never be permanently removed and must not be reasoned about as though they could be. Mark them once as recurring, and thereafter the only question is "is the condition true right now?" — never "how do we solve this?".
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- yesterday Changed · +11 lines 604747a5b33c
- 5d ago First seen · 90 lines · 132 tokens per session scan A b6defa0c77a5
ungate-queue is a skill published in the GitHub repository haiggoh/run-to-completion (1 stars, last pushed 2d ago), licensed MIT. It adds 132 tokens to every session and 2,215 once invoked, about $0.0007 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
extracting-requirements
Use when starting an iterative-development run on human spec collateral — reads the spec, produces per-epic requirement files with proof obligations and behavior scenario cards with stable IDs.
iterative-development
Use when implementing a project with a large, comprehensive, or ambiguous spec — extracts requirements with proof obligations, defines a walking skeleton with its first journey scenario, then loops through audited sprints that continuously build a behavior evidence corpus. Completion means passing evidence, not just…
scoping-the-simplest-core
Use when turning extracted requirements into a roadmap — selects the walking skeleton iteration with its first journey scenario, orders remaining work into follow-on iterations, and applies story splitting when ACs have different dependency profiles.
autogpt-agents
Autonomous AI agent platform for building and deploying continuous agents. Use when creating visual workflow agents, deploying persistent autonomous agents, or building complex multi-step AI automation systems.
prd
Create a full Product Requirements Document (PRD-NNN) via a structured PM interview — for large initiatives that justify the overhead. Use when PM mentions "create PRD", "write product requirements", "PRD document", "product brief", "requirements doc", "напиши PRD", "создай PRD", or any request to capture a full…
review-pr
Run a quality review on an open pull request (by PR number or linked TASK), externally via codex or internally via a clean-context self-review, and post the verdict. Use when PM mentions "review PR", "PR review", "review pull request", "quality review", "review this pr", "сделай ревью pr", or any request to evaluate…