CA wei-cross-research
Cross-validate research answers by querying multiple LLMs in parallel with judge-based synthesis. Reduces hallucination and surfaces model disagreements for high-stakes questions. 交叉研究：多模型并行查询与裁判合成，通过交叉验证降低幻觉、暴露分歧，适用于高 stakes 决策。
As a process A 88/100 · Runs to the end — weak spots: running it twice
What is at stake
The skill contains fragments that, in the wrong hands, cost money or data. Below: what the installer risks and what the author should do.
The text contains phrases like "ignore previous instructions" or "you are now…". That is an attempt to hijack the agent: it may break your rules, the system limits or company policy.
An honest skill does not need them: state the role and the rules directly without overriding other instructions. Otherwise catalog scanners and corporate filters will block the listing.
Below is the worst case for this category. The finding here is medium: the guard saw a sign, not a proof.
The instructions or scripts send data out: environment variables, keys, file contents, chat history. You may never notice, because the agent performs the upload as "part of the task".
If the upload is not needed for the task, remove it: catalogs flag such skills and delist them. If it is needed, name the destination explicitly, say what leaves the machine, and give the user a switch.
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.
How to improve
- Address the high-severity findings: each costs 18 safety points. If one is a false positive, add the rule id to guard.allow in spec.yaml.
- 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 · 11
-
high Instruction override
en-ignore-previousscripts/agent.ts:286Instruction-override phrase ("ignore previous instructions")/ignore previous instructions/gi,
Medium and low: 10
-
medium Exfiltration
net-redirectable-api-keyscripts/clients/bailian.ts:152Helper sends the API key to a host configured by an environment variable — the key can be redirected to another serverAPI key + configurable base URL from environment
-
medium Exfiltration
net-redirectable-api-keyscripts/clients/openai_compliant.ts:178Helper sends the API key to a host configured by an environment variable — the key can be redirected to another serverAPI key + configurable base URL from environment
-
medium Exfiltration
net-redirectable-api-keyscripts/clients/openrouter.ts:132Helper sends the API key to a host configured by an environment variable — the key can be redirected to another serverAPI key + configurable base URL from environment
-
medium Dangerous commands
cmd-pipe-to-shell-known-hostSKILL.md:84Pipe-to-shell installer from a well-known host (still executes remote code)curl -fsSL https://bun.sh/install | bash
-
low Secrets in code
secret-high-entropy-tokenpackage-lock.json:26High-entropy token-like string (may be an id, hash or a credential) (detector / deny-list definition)"integrity": "sha5…7pY+zoMV…h0x/Ptw8…8dg==",
detector -
low Secrets in code
secret-high-entropy-tokenpackage-lock.json:43High-entropy token-like string (may be an id, hash or a credential) (detector / deny-list definition)"integrity": "sha5…b00+Gxjx…zRc/oZwU…hzA==",
detector -
low Secrets in code
secret-high-entropy-tokenpackage-lock.json:111High-entropy token-like string (may be an id, hash or a credential) (detector / deny-list definition)"integrity": "sha512-vHk/hA7/1Ack…wbo+jaSh…Gtl+A5zq…HFg==",
detector -
low Secrets in code
secret-high-entropy-tokenpackage-lock.json:145High-entropy token-like string (may be an id, hash or a credential) (detector / deny-list definition)"integrity": "sha5…Dsc+j03S…0oA==",
detector -
low Secrets in code
secret-high-entropy-tokenpackage-lock.json:434High-entropy token-like string (may be an id, hash or a credential) (detector / deny-list definition)"integrity": "sha5…NS8+tHW7…WOF+PEzk…X4Q==",
detector -
low Instruction override
en-ignore-previousscripts/agent.ts:265Instruction-override phrase ("ignore previous instructions") (detector / deny-list definition; code comment; security demo / example)* SECURITY NOTE: The patterns below (e.g., "ignore previous instructions") are
detectorcommentdemo
Files scanned: 20. Evidence is masked. Grey chips explain why severity was lowered.
Against the Agent Skills spec
- note
frontmatter-keyunknown frontmatter key "execution" - note
frontmatter-keyunknown frontmatter key "env"
Process rating: all ten parameters 88/100
- 30Running it twice. 2 mutating operations with no state check
- 60Result and completion. Output format stated, no completion criterion
- 70Inputs and preconditions. Inputs and preconditions are listed
- 100Tools and files. No external tools needed
- 100Steps. 57 steps
- 100When it triggers. States when to use and when not to
- 100Failures and branches. 4 branches, has a failure section
- 100Consistency. Name and required fields are in place
- 100Execution cost. Instruction body is 3798 tokens
- 100Progress reporting. Reports progress
- medium Safety rules and hard prohibitions inside a skill: they belong in the system prompt, here they protect nothing
- low 15 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
- -31 of 2 scripts are never mentioned in SKILL.md
- +1No license
- +2Single-language instructions
- +4Description says when NOT to use the skill
- +3Description length 229: enough signal without eating the budget
- +4Structure: 38 headings
- +3Step-by-step instructions: 57 items
- +3Output format is stated explicitly
- +4Has examples (15 code blocks)
Quality base 70; lint remarks subtract, signals add up to 100. Result: 86.