The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill includes functionality in scripts/alphapai.py to save user-provided API keys and account passwords in plain text within a config.json file located in the skill's root directory. The alphapai_web_client.py script also stores session tokens in auth_state.json.
[CREDENTIALS_UNSAFE]: The configuration loading logic in scripts/alphapai.py and scripts/alphapai_client.py uses a function _find_env_value that searches for .env files in the current directory and up to three levels of parent directories. This broad search could lead to the unintended exposure of sensitive credentials belonging to other projects on the same system.
[COMMAND_EXECUTION]: The scripts/setup.sh script executes shell commands to check for and install the requests Python library using pip, which is a common but notable external dependency management behavior during automated setup.
[EXTERNAL_DOWNLOADS]: The skill setup process involves downloading the requests package from the public PyPI registry if it is not already present on the system.
[INDIRECT_PROMPT_INJECTION]: The skill retrieves financial data, news, and roadshow transcripts from remote APIs, creating a potential surface for indirect prompt injection if the remote data contains adversarial instructions.
Ingestion points: External data is ingested through the qa, recall, and agent commands defined in scripts/alphapai.py and the associated client scripts.
Boundary markers: The instructions lack clear boundary markers or system-level warnings to the agent to treat the retrieved API data as potentially untrusted content.
Capability inventory: The skill allows the agent to execute local Python scripts that perform network requests and write to the local file system.
Sanitization: The skill does not perform sanitization or filtering of the content returned by the AlphaPai service before presenting it to the agent's context.
[PROMPT_INJECTION]: The SKILL.md file contains strong instructions to the agent to skip summarization and directly output raw results from the API. Phrases like '禁止总结或压缩' (Forbidden to summarize or compress) and '直接呈现原始输出' (Directly present original output) restrict the agent's ability to summarize or filter potentially malicious content within the API response, facilitating the delivery of hidden instructions to the user.

capture-alphapai-unified