BC Sentinel — Agent Security Layer
Runtime security layer for OpenClaw agents. Intercepts and scans all external input (emails, API responses, web content, chat messages, calendar events) for prompt injection, data exfiltration, credential leaks, and social engineering BEFORE the agent processes it. Also monitors agent output for secret leakage and suspicious command requests. Use when: your agent processes untrusted external data, you need automatic input sanitization, output monitoring to prevent data leaks, or multi-language injection detection (EN/RU/ZH/ES/AR).
As a process C 60/100 · Has gaps — weak spots: when it triggers, failures and branches, consistency
What is at stake
Medium-severity findings: the skill is probably honest, but read what alarmed the scanner.
Below is the worst case for this category. The finding here is medium: the guard saw a sign, not a proof.
The skill contains commands that delete files, rewrite disks or execute code fetched from the network. The agent may run them without asking if it believes the instructions require it.
Replace destructive commands with safe equivalents that ask for confirmation, scope them to one folder, and stop piping curl into a shell: pin a version and a checksum.
Below is the worst case for this category. The finding here is medium: the guard saw a sign, not a proof.
The purpose itself is risky: wallets, browser password stores, offensive security. Even an honest implementation gives the agent access to things that cost money.
Explain in the description why the access is needed and how it is limited; add tests that show refusals on dangerous requests.
How to improve
- Your own cases (evals/evals.json, 4–6 real requests with expected answers): the full check would then run those instead of a model-drafted suite.
- A spec.yaml with trigger phrases and assertions — a behaviour contract for CI; `skilltest init` writes a template.
Guard findings · 6
✓ No critical or high findings
Medium and low: 6
-
medium Dangerous commands
cmd-encoded-execscripts/sentinel-input.sh:257Executes a base64/encoded payload (string literal in code, not executed; documentation of a security skill)"base64 -d.*\| ?(bash|sh|python|perl|ruby)"
code literalsecurity skill -
medium Risky intent
intent-wallet-secretsscripts/sentinel-output.sh:194Handles crypto-wallet secrets (seed / mnemonic / private key) — a classic stealer target# BIP-39 Seed Phrase detection (12 or 24 common BIP-39 words in sequence)
-
low Risky intent
intent-wallet-secretsscripts/sentinel-output.sh:201Handles crypto-wallet secrets (seed / mnemonic / private key) — a classic stealer target (detector / deny-list definition)THREATS+=("crypto_leak: Possible BIP-39 seed phrase ($SEED_MATCH/12+ matching words)")detector -
low Secrets in code
secret-high-entropy-tokenSKILL.md:159High-entropy token-like string (may be an id, hash or a credential) (quoted — discussed, not commanded)- 🪙 Tip (USDT TRC-20): `TMkk…8Y3`
quoted
A further 2 matches are quotations in this security skill's documentation and are not counted as findings.
Files scanned: 8. Evidence is masked. Grey chips explain why severity was lowered.
Against the Agent Skills spec
- warning
name-formatname should be kebab-case (lowercase letters, digits, hyphens)
Process rating: all ten parameters 60/100
- 0Failures and branches. Linear process with no failure handling
- 20When it triggers. No condition that starts the skill
- 40Consistency. Frontmatter name (Sentinel — Agent Security Layer) differs from the folder (sentinel-oleg)
- 60Tools and files. Uses tools (bash, web) that frontmatter does not declare
- 60Result and completion. Output format stated, no completion criterion
- 70Inputs and preconditions. Inputs and preconditions are listed
- 100Steps. 20 steps
- 100Execution cost. Instruction body is 1086 tokens
- 100Running it twice. No mutating operations
- 100Progress reporting. Reports progress
- low 10 top-level sections: this looks like several domains in one skill
Everything here is measured from the skill text rather than judged by a model, so the numbers are checkable. A parameter weighs more when it is a more common reason for the process to stall.
Quality signals
- +5Description has no quoted example phrases that should trigger the skill
- +4Description does not say when NOT to use the skill (false activations)
- -221 emoji in the instructions: noise for the model
- -31 of 5 scripts are never mentioned in SKILL.md
- +2Single-language instructions
- +3Description length 536: enough signal without eating the budget
- +4Structure: 16 headings
- +3Step-by-step instructions: 20 items
- +3Output format is stated explicitly
- +4Has examples (8 code blocks)
- +1License stated
Quality base 70; lint remarks subtract, signals add up to 100. Result: 78.