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 zero-lab-ai/skills --skill covert-channel-assessmentgit clone --depth 1 https://github.com/zero-lab-ai/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/zero-lab-ai/skills/covert-channel-assessment)<a href="https://agentmods.dev/skills/zero-lab-ai/skills/covert-channel-assessment"><img src="https://agentmods.dev/badge/skills/zero-lab-ai/skills/covert-channel-assessment/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/zero-lab-ai/skills/covert-channel-assessment"><img src="https://agentmods.dev/badge/skills/zero-lab-ai/skills/covert-channel-assessment.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 | $0.00066 | $0.01212 |
| Opus 5.5 | $0.00026 | $0.00485 |
| Sonnet 5.5 | $0.00013 | $0.00242 |
| Haiku 4.5 | $0.00007 | $0.00121 |
Grade A, and why
covert-channel-assessment 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.
Covert channel assessment
A permitted protocol is not a working covert channel. A working channel is not automatically undetected. Keep route feasibility, execution proof, detection outcome, and data custody as separate claims.
This skill assesses supplied evidence and bounded synthetic tests. It does not provide payloads, encodings, tunneling commands, evasion settings, C2 design, pivoting, persistence, credential use, third-party abuse, or authority to move real data.
When to Use
- Evaluating DNS, web, cloud, ICMP, tunnel, timing, steganographic, or dead-drop paths without turning a plausible route into a claim of successful exfiltration.
- Use only within explicit authorization and the stated rules of engagement.
When NOT to Use
- Do not treat this skill as permission to access, test, or change a target.
- Do not convert missing evidence into a clean result or a stronger claim.
Workflow
- Record the written authorization, exact sources, assessment window, approved receivers, synthetic marker hash, byte and duration ceilings, prohibited data, services, and actions. Receiver ownership and authorization are different facts.
- Inventory evidence with source, locator, honest time precision, and integrity hash for copied artifacts. Use null time fields when the source supplies no observation time; the assessment window is not evidence time. Preserve unknowns; do not invent throughput, queries, alerts, endpoints, accounts, transfer times, or successful delivery.
- For every source, assess all eight approved channel families. Keep a family unknown or blocked when evidence is absent or testing is prohibited.
- Treat
plausibleas a route hypothesis. Usevalidatedonly after an authorized test records the approved marker, receiver, exact bounded times, observed bytes, source evidence, and receiver-side delivery evidence. - Describe capacity as
unknown,estimated, ormeasured. Only an executed test can produce measured throughput. Baselines describe normal traffic, not covert channel capacity. - Inventory telemetry by source, health, retained fields, window, and limitation. Packet counts do not prove payload inspection; process telemetry does not prove delivery; a detection rule does not prove an alert fired.
- Predeclare a detection check for every planned or executed test.
not-detectedrequires healthy telemetry queried across the entire test window. An absent alert without that evidence isinconclusive. - Enforce byte and duration ceilings per test and in aggregate per source. Keep blocked or prohibited paths visible; do not force a transfer to complete coverage.
- Track marker creation, source staging, receiver storage, retention, and destruction for every planned or executed test. A synthetic marker still has a custody trail.
- Link medium-or-higher channels to an owned response: validate safely, improve telemetry, tighten egress, or accept the documented risk.
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.
- yesterday First seen · 101 lines · 66 tokens per session scan A b9052a754302
covert-channel-assessment is a skill published in the GitHub repository zero-lab-ai/skills (283 stars, last pushed 12d ago), licensed Apache-2.0. It adds 66 tokens to every session and 1,212 once invoked, about $0.0003 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 2026-10-26.
Other skills, from other repositories
performing-ai-assisted-vulnerability-discovery
Using LLMs to accelerate vulnerability research and pentest workflows — generating syntax-valid fuzzing seeds and evolving grammars, fine-tuned mutation dictionaries, parallel agent-based proof-of-vulnerability generation, and evidence-driven passive analysis of real HTTP traffic via the Burp MCP server. Covers…
testing-llm-prompt-injection-and-jailbreaks
Testing LLM-backed applications, chatbots, and AI agents for direct and indirect prompt injection, jailbreaks, system-prompt leakage, and tool/agent abuse during authorized penetration tests, using structured payload families and reliable confirmation signals.
bypassing-authentication-with-forced-browsing
Discovering and accessing unprotected pages, APIs, and administrative interfaces by enumerating URLs and bypassing authentication controls during authorized security assessments.
analyzing-ios-app-security-with-objection
Performs runtime mobile security exploration of iOS applications using Objection, a Frida-powered toolkit that enables security testers to interact with app internals without jailbreaking. Use when assessing iOS app security posture, bypassing client-side protections, dumping keychain items, inspecting filesystem…
conducting-full-scope-red-team-engagement
Plan and execute a comprehensive red team engagement covering reconnaissance through post-exploitation using MITRE ATT&CK-aligned TTPs to evaluate an organization's detection and response capabilities.
exploiting-ai-model-file-rce
Testing machine-learning model files and model-loading services for remote code execution caused by insecure deserialization (pickle/PyTorch), unsafe config instantiation (Hydra), archive path traversal, and dangerous layer types during authorized penetration tests of AI/ML pipelines.