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 boshu2/agentops --skill one-way-doorgit clone --depth 1 https://github.com/boshu2/agentopsWrote 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/boshu2/agentops/one-way-door)<a href="https://agentmods.dev/skills/boshu2/agentops/one-way-door"><img src="https://agentmods.dev/badge/skills/boshu2/agentops/one-way-door.svg" alt="Measured on agentmods" height="20"></a>- NVIDIA SkillSpector warn
SkillSpector: 3 findings, up to high
These are SkillSpector’s own severities. On a checked sample its high-severity flags on skills were ~96% false positives — a documented command, a public API, a “never do X” rule — so we show them as a caution to read, not a verdict. Why →
- high Tool Misuse · line 158 Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).Fix: Validate all tool parameters against an allowlist. Reject dangerous parameter values (shell=True, --force, -rf /) and use safe defaults.
- medium Excessive Agency · line 40 Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.Fix: Add human-in-the-loop confirmation for destructive, irreversible, or high-impact operations. Never auto-execute commands that modify files, send data, or alter system state.
- medium Agent Snooping · line 188 Skill enumerates or reads other installed skills. Access to other skills' SKILL.md files or the skills directory reveals prompt instructions, capabilities, and secrets that should be invisible to peer skills.Fix: Remove all code or instructions that list or read other skills' files or directories. Skills should operate independently; cross-skill access is a privilege escalation.
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 | $0.00073 | $0.02061 |
| Opus 5 | $0.00036 | $0.01030 |
| Sonnet 5 | $0.00015 | $0.00412 |
| Haiku 4.5 | $0.00007 | $0.00206 |
Grade A, and why
one-way-door 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 4d 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 — 199 lines — stays where its author put it; the contents beside it link to each section on GitHub.
/one-way-door
Classify one pending decision as two-way (cheap to undo, may be auto-decided) or one-way (expensive or impossible to undo, always surfaced to the caller). Return the classification and stop. This skill decides who decides. It never decides the underlying question and it never performs the action.
Insight: an agent's confidence is uncorrelated with a decision's reversibility, so confidence is the wrong gate. Reversibility is a property of the decision itself, is knowable before acting, and can be declared in advance instead of inferred from how the question happened to be worded.
The failure mode this exists to prevent: an agent auto-approves an irreversible step because the surrounding batch of decisions was routine and the step read as one more item in the list. Dropped table, force-push over a colleague's work, rotated credential, published post. The batch is why it happens: nineteen two-way decisions train the reflex that answers the twentieth.
dcg guards destructive commands at the shell boundary. This guards
destructive decisions at the judgment boundary, before any command exists.
Modes
| Trigger phrases | Mode | Entry point |
|---|---|---|
| "is this a one-way door", "can we undo this", "classify this decision" | classify | the four layers below |
| "should I just decide this", "auto-decide the rest" | gate a batch | classify each, act only on two-way |
| "the models disagree with me", "both reviewers say I'm wrong" | user challenge | the packet below |
Inputs
Required: a one-line summary of the pending decision, and the effect it would have if answered wrong.
Optional: a declared decision id from references/decision-registry.md, and the
skill or lane raising it.
Non-goals. This skill does not answer the decision, rank options, execute
anything, or record the caller's answer. It does not gate shell commands — that
is dcg. It does not model caller preferences or build a profile of how the
caller usually answers; a preference that suppresses questions is exactly the
mechanism that must never reach a one-way door.
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.
- 4d ago Changed · +9 lines · +2 tokens per session 91e28caab8a0
- 8d ago First seen · 190 lines · 71 tokens per session scan A fada7b6eb9c0
one-way-door is a skill published in the GitHub repository boshu2/agentops (433 stars, last pushed 2d ago), licensed Apache-2.0. It adds 73 tokens to every session and 2,061 once invoked, about $0.0004 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 2026-08-30.
Other skills, from other repositories
brainstorm
Explore vague or ambitious ideas into a right-sized requirements-only plan. Use when the user wants to brainstorm, think through scope, decide what to build, or needs collaborative product framing before planning, not for a decisive verdict on whether to adopt or switch to a specific external technology, library, or…
audit-project
Run an iterative multi-agent code audit until critical and high findings are resolved. Use when the user says "audit my code", "find all the bugs", "deep code audit", "iterative review", or "review until clean".
commit-push-pr
Use when asked to commit, push a feature branch, and open or update a pull request with gh in one pass. Not for a push with no PR: use commit-push. Not for a PR body alone: use create-pull-request.
commit
Use when asked to commit changes, create a typed branch, format history for a changelog, or rewrite messages of HEAD or an unpushed range. Not for pushing or a PR: use commit-push or commit-push-pr.
drift-detect
Use when the user says "plan drift", asks whether the roadmap, plans, or docs still match the code, or is deciding what to rebuild when restarting a stalled project. For doc-vs-code drift inside a specific diff, use sync-docs.
duet
Use when the user invokes /duet, says "pair with me", or faces aesthetic, architectural, or irreversible decisions.