The Agent Skills Directory

[DYNAMIC_EXECUTION]: scripts/content_ingester.py contains a self-management routine that automatically installs missing packages from requirements.txt using pip install and then uses os.execv to perform process replacement, restarting the script with the newly installed environment.
[COMMAND_EXECUTION]: The skill frequently uses subprocess.run and subprocess.check_call to chain internal scripts and manage system-level operations, such as dependency installation and triggering the generation pipeline.
[DYNAMIC_EXECUTION]: scripts/mentor_relay.py starts a local HTTP server on 127.0.0.1:8760. This relay acts as a proxy to allow interactive HTML documents (which are opened locally in a browser) to send API requests to AI models without triggering CORS restrictions.
[DATA_EXFILTRATION]: The skill collects text and metadata from user-provided files and URLs and transmits this data, along with API keys (such as DASHSCOPE_API_KEY) retrieved from environment variables, to external cloud services including Alibaba DashScope and other user-specified OpenAI-compatible endpoints.
[INDIRECT_PROMPT_INJECTION]: The skill is highly exposed to indirect prompt injection due to its primary purpose of ingesting untrusted external content.
Ingestion points: scripts/content_ingester.py reads data from remote URLs (via DrissionPage or requests) and various file formats (PDF, Word, Images via OCR).
Boundary markers: The skill uses 'Truth Anchoring' logic in scripts/truth_anchor.py and structured output contracts defined in references/system_prompt.md to attempt to distinguish between source data and agent instructions.
Capability inventory: The agent has capabilities for file writing, network requests to external APIs, and shell command execution via subprocess.
Sanitization: scripts/content_ingester.py performs basic sanitization by stripping script/style tags and removing UI-related 'noise' patterns from ingested HTML.

knowledge-absorber