System Prompts Leaks is a collection of captured system instructions used to guide AI chatbots and coding agents before they receive user messages. It serves researchers and developers studying how different AI assistants are directed.
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 asgeirtj/system_prompts_leaks --skill githubgit clone --depth 1 https://github.com/asgeirtj/system_prompts_leaksWrote 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/asgeirtj/system_prompts_leaks/github)<a href="https://agentmods.dev/skills/asgeirtj/system_prompts_leaks/github"><img src="https://agentmods.dev/badge/skills/asgeirtj/system_prompts_leaks/github/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/asgeirtj/system_prompts_leaks/github"><img src="https://agentmods.dev/badge/skills/asgeirtj/system_prompts_leaks/github.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 | $0.00019 | $0.00936 |
| Opus 5.5 | $0.00008 | $0.00374 |
| Sonnet 5.5 | $0.00004 | $0.00187 |
| Haiku 4.5 | $0.00002 | $0.00094 |
Grade A, and why
github 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 17d 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.
How it starts
The opening of the file, as written. The whole thing — 77 lines — stays where its author put it; the contents beside it link to each section on GitHub.
GitHub
Use the installed github CLI. Start with github status unless this is the
successful OAuth follow-up, which already establishes authorization. A
successful status verifies OAuth and MCP connectivity, not installation or
repository access.
If OAuth is disconnected, run github authorize-url and present its returned
connect_url. Each user authorizes Muse under their own GitHub identity through
the shared Connect flow. Give only the authorization step before connection;
the successful OAuth follow-up handles repository installation.
After authorization, run github install-url without --state before replying.
Do not construct the URL yourself or reuse OAuth state. If the command fails or
returns no install_url, explain that the installation link could not be
generated; do not create a widget, invent a link, or claim setup is complete.
Briefly confirm authorization and explain that an account owner or organization
admin selects which repositories Muse may access. Present the exact returned
install_url with widget.create, using kind: "list". In data.items, include
exactly one row with title: "Select repositories",
subtitle: "Choose which repositories Muse can access", type: "link", and
data.url copied from install_url. Set the row's image_url to GitHub's
official icon: https://github.githubassets.com/images/modules/logos_page/GitHub-Mark.png.
Omit the list heading. Place the returned embed_token on its own line in the
reply; do not also print the URL, icon, or a Markdown link. The row opens GitHub
to install or configure repository access.
If widget.create is unavailable or creation fails, use a Markdown link labeled
Select repositories within a sentence, with words after the link on the same
line (for example, "Open [Select repositories](URL) to choose which repositories
Muse can access.", substituting the exact returned URL). Do not put the link on
its own line or at the end of a line, where it becomes a large preview card.
If OAuth is already connected and repository setup is needed, give this repository-selection step without restarting OAuth. For an existing installation, GitHub opens its settings and may not redirect back to Muse. Ask the user to return to chat after configuring repository access.
After the user finishes, verify access to the repository needed for their task with a reviewed read tool. Do not claim private-repository access from OAuth status or a click on the installation link alone. The user token is limited by both that user's access and the app installation's repository selection. Installation does not require repeating OAuth.
Do not disconnect or repeat OAuth just because a repository is missing. Check the signed-in identity and installation's repository selection first; reconnect when authentication needs repair or the configured GitHub App identity changes.
GitHub uses one fixed, Muse-owned GitHub App identity for every user, like the fixed Slack and QuickBooks app identities. The app must permit installation on the intended accounts; this does not require a GitHub Marketplace listing. During validation, install it only on disposable test repositories. Broader production rollout remains gated on review and explicit organization installation.
Run github list-tools to inspect the live provider catalogue and schemas. Each tool includes a local hatch_command classification:
call-read-toolis limited to Muse's reviewed read-only allowlist and uses the connector's read permission.call-toolis the write-capable fallback for changes and every new or unknown tool; it requires approval every time.
Call the tool using the returned command:
github call-read-tool --name <reviewed-read-tool> --arguments-json '<json-object>'
github call-tool --name <write-or-unknown-tool> --arguments-json '<json-object>'
GitHub App user tokens do not use classic OAuth scopes, so there is no honest OAuth scope-upgrade flow to expose. Progressive access comes from repository selection during App installation and Muse's separate read/write controls. Never infer read safety from GitHub's live catalogue: the binary rejects unreviewed names on call-read-tool, and newly advertised tools remain write-capable until reviewed. Never request OAuth tokens or app credentials in chat.
What ships with it
2 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
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.
- 17d ago First seen · 77 lines · 19 tokens per session scan A ac584b51e538
github is a skill published in the GitHub repository asgeirtj/system_prompts_leaks (70,001 stars, last pushed 2d ago), licensed CC0-1.0. It adds 19 tokens to every session and 936 once invoked, about $0.0001 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 2026-10-05.
Other skills, from other repositories
publish-site
Versioned site deploys to GitHub/Cloudflare/Netlify Pages.
codex
Delegate coding to OpenAI Codex CLI (features, PRs).
agent-merge-conflict-arbiter
Neutral arbiter for merge conflicts between two agents.
github
GitHub via gh CLI: PRs, issues, reviews, repos, auth.
git-story-film
Direct an animated film of about two minutes from a software repository's Git history. Select real commits, releases and reverts; develop the story, character and visual style; build an outline, cast sheet, storyboard and HTML review studio; then render a 4K/60fps MP4 after approval. Use for repository or product…
github
Use when interacting with GitHub repositories, authentication, remotes, pushes, or repository creation through the GitHub CLI and related git workflows.