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 skills add zebbern/agent-collab --skill goal-runnergit clone --depth 1 https://github.com/zebbern/agent-collabWrote 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/zebbern/agent-collab/goal-runner)<a href="https://agentmods.dev/skills/zebbern/agent-collab/goal-runner"><img src="https://agentmods.dev/badge/skills/zebbern/agent-collab/goal-runner/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/zebbern/agent-collab/goal-runner"><img src="https://agentmods.dev/badge/skills/zebbern/agent-collab/goal-runner.svg" alt="Reviewed on agentmods" width="80" 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 | $0.00062 | $0.00605 |
| Opus 5.5 | $0.00025 | $0.00242 |
| Sonnet 5.5 | $0.00012 | $0.00121 |
| Haiku 4.5 | $0.00006 | $0.00060 |
Grade A, and why
goal-runner 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 28d ago.
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.
What it actually says
Goal Runner
The companion (goal-companion.mjs) owns state, selection, and dispositions;
you own judgment. Follow the /goal:step choreography exactly — never
reimplement its bookkeeping by editing the goal file ad hoc.
Policy:
- One increment at a time.
next→start→ execute → verify → PR →record→ stop. The companion refuses a second in-progress item; do not work around the refusal. - Disclose. Announce each increment (and each delegation) in one line.
- Start before working. Call
startbefore any work on the increment begins — including delegated work — or the ledger's step-started→disposition gap understates reality and the retro must discard it. Ledger evidence, 2026-08-07: one of eight durations read 0min because work precededstart. - Delegate with judgment. Trivial work stays local. Deep analysis or a
second opinion goes to Codex, fast parallel implementation to Cursor, via
the
codex-delegation/cursor-delegationskills. Analysis and implementation are separate delegations when both are needed. - Refine once. If delegated work fails verification, refine the brief
with the failure evidence and re-delegate once; then record
blocked. - Blocked is a full stop. A blocked item blocks the goal;
nextrefuses until a human resolves it. Surface the reason; never guess past it. - Unattended never merges. A
/loop-driven or scheduled step works on agoal/<slug>/<itemId>branch, opens the PR, leaves the item in-progress, and stops; the next wake reconciles that PR (merged / closed / still open) using/goal:step's all-state JSON query and exact head-branch check. Failed, missing, or ambiguous PR evidence stops the wake without a disposition. Recording a reconciliation also ends the wake. Merging stays human. - One writer per project. Prevent overlapping scheduled wakes and manual
/goal:stepinvocations. If another runner may still be active, stop before changing goal state. - Honest dispositions.
mergedmeans the PR merged. Never invent progress, never record a disposition that has not actually happened, and if the companion or a delegate fails, report that instead of substituting your own output.
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.
- 28d ago Changed · +6 lines 6096cd618bcc
- 1mo ago First seen · 38 lines · 62 tokens per session scan A e58cd4f017c1
goal-runner is a skill published in the GitHub repository zebbern/agent-collab (34 stars, last pushed 28d ago), licensed Apache-2.0. It adds 62 tokens to every session and 605 once invoked, about $0.0002 per session on Opus 5.5. 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 2026-08-30.
Other skills, from other repositories
skill-builder
Automatically detect source types and build AI skills using Skill Seekers. Use when the user wants to create skills from documentation, repos, PDFs, videos, or other knowledge sources.
ask-agents
Send a specific question or task to the external agent or agents the user names and relay their responses. Use for an outside opinion or explicit request to ask Claude, Codex, or Gemini.
ai-ml-development
AI and machine learning development with PyTorch, TensorFlow, and LLM integration. Use when building ML models, training pipelines, fine-tuning LLMs, or implementing AI features.
android-development
Android development with Kotlin, Jetpack Compose, and modern Android architecture. Use when building Android apps, implementing Material Design, or following Android best practices.
email-systems
Transactional email (Resend, SendGrid, SES), templates (React Email, MJML), deliverability (SPF/DKIM/DMARC), and inboxing best practices. Use when building email infrastructure, designing templates, or troubleshooting deliverability.
agent-svelte-expert
Specialist subagent: Svelte 5 runes, SvelteKit routing, form actions, and compile-time reactivity specialist. Use when building Svelte or SvelteKit applications, implementing rune-based state management, or working with SvelteKit features. Trigger phrases: Svelte, SvelteKit, runes, $state, $derived, $effect, $props…