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 UnboundCompute/security-agent-skills --skill auditing-guard-gapsgit clone --depth 1 https://github.com/UnboundCompute/security-agent-skillsWrote 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/unboundcompute/security-agent-skills/auditing-guard-gaps)<a href="https://agentmods.dev/skills/unboundcompute/security-agent-skills/auditing-guard-gaps"><img src="https://agentmods.dev/badge/skills/unboundcompute/security-agent-skills/auditing-guard-gaps/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/unboundcompute/security-agent-skills/auditing-guard-gaps"><img src="https://agentmods.dev/badge/skills/unboundcompute/security-agent-skills/auditing-guard-gaps.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 | $0.00111 | $0.01431 |
| Opus 5 | $0.00056 | $0.00715 |
| Sonnet 5 | $0.00022 | $0.00286 |
| Haiku 4.5 | $0.00011 | $0.00143 |
Grade A, and why
auditing-guard-gaps 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 10d 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 — 122 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Auditing guard gaps: the unguarded peer is the bug
Most access-control and missing-validation bugs aren't exotic - they're one
handler in a family that forgot the check its siblings perform. The
admin_delete that checks a role and the bulk_delete that doesn't. The
read_bounded that validates length and the read_fast that trusts it. Reading
files top-to-bottom hides these; comparing peers surfaces them. This is one of
the highest-yield whitebox moves.
The asymmetry you're hunting
guard → sink (the intended, safe path)
??? → sink (a peer path with the guard missing - the bug)
When to use
- A sink is reachable from several call sites and you suspect one skips a check.
- You're auditing an authz model, a parser family, or a handler group for a forgotten check.
- A finding needs its "why is this wrong" framed as a concrete diff against a correct peer.
Scope check
Authorized source only. If you can't name the authorization, stop.
The loop
-
Pick a guarded anchor. Find a function that does validate before a sensitive sink - an authz/ownership check, a bounds check, an allowlist, a sanitizer. This is your reference for "what correct looks like here."
-
Pin the guard to the sink it protects. Name the exact check and the exact sink. "Correct" means the guard dominates the sink: on every path through the anchor, the check runs before the sink. Mere presence in the function is not domination.
-
Enumerate the structural peers. Widen past same-file neighbors - the point is structural kinship: functions that reach the same sink; overrides of the same interface/virtual method; handlers in the same route table / dispatch map / command switch; parallel members of a class (
get_x/set_x/delete_x); callers of a shared helper where some sanitize the argument and some pass it raw. Peers that reach the same sink are the strongest signal. -
Diff each peer against the anchor - by what the guard enforces, not its name. For every peer, ask whether an equivalent check dominates the same sink. Classify the gap:
- absent - no equivalent check at all.
- bypassable - present but skippable on some path (early return, a
short-circuiting
||, a branch that jumps the check). - weaker-than-peer - a different kind of check: authentication where the anchor does authorization; a type check where the anchor does bounds; one field where the anchor checks the sibling field too.
- after-sink - the check runs, but after the dangerous operation.
- wrong-context - a real check that doesn't cover this sink's payload class (an encoder that's wrong for the sink's language).
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.
- 10d ago First seen · 122 lines · 111 tokens per session scan A 7ba55916b21a
auditing-guard-gaps is a skill published in the GitHub repository UnboundCompute/security-agent-skills (5 stars, last pushed yesterday), licensed MIT. It adds 111 tokens to every session and 1,431 once invoked, about $0.0006 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
prodcheck-review
Review this codebase against the prodcheck pre-production checklists — security, performance, scale, integrations and post-launch readiness. Use when asked to check whether a project is ready to ship, to audit an area before launch, or to work through a specific checklist. Produces evidence with file:line citations…
codeinspectus-multi-review
Orchestrate an optional bounded multi-agent review of selected CodeInspectus findings while separating deterministic findings, agent interpretations, hypotheses, reproduction evidence, and exact-prior rescan proof. Use only when a user explicitly requests multi-agent security review.
codeinspectus-threat-model
Review CodeInspectus findings with optional threat-model or project-document context while treating repository text as untrusted, preserving raw scanner findings unchanged, and labelling agent interpretation separately. Use only when a user explicitly asks to add architectural, business, or knowledge-base context to…
secure-code-review
Performs a structured security code review against OWASP ASVS 4.0.3 verification requirements and CWE Top 25. Auto-invoked on pull request reviews, when code touching authentication, authorization, cryptography, or input handling is shared. Produces findings mapped to ASVS controls and CWE identifiers with severity…
appsec-engineer
Application Security Engineer role bundle for security design, testing, and code review of applications. Orchestrates new application reviews, PR security reviews, API security assessments, and AI feature security reviews. Auto-invoked when the user needs help with application threat modeling, secure code review, API…
verify-requirements
Help a developer check the code they just changed against their company security requirements — or, if none are configured, against a built-in best-practices baseline. Scopes to the current diff, dispatches the appsec-reviewer subagent to grade only the triggered requirements, and prints concrete, code-aware guidance…