The Agent Skills Directory

[COMMAND_EXECUTION]: The reference files include numerous manual diagnostic commands intended for the developer to use for tree verification. These commands utilize standard utilities like grep, find, sed, awk, unzip, and javap to audit source code and build artifacts for specific patterns or misconfigurations.
[CREDENTIALS_UNSAFE]: The documentation provides guidance on the secure handling of API keys and proprietary credentials in multiplatform projects (e.g., foss-vs-proprietary-module-pairs.md). It explicitly advocates for best practices such as utilizing gradle.properties, local.properties, or the buildkonfig plugin to manage secrets externally rather than hardcoding them in the source code.
[INDIRECT_PROMPT_INJECTION]: The skill describes patterns for ingesting and processing untrusted external data, such as HTML entity decoding (kmp-html-entity-decoder.md), fuzzy matching (levenshtein-fuzzy-match-pure-kmp.md), and timestamped token parsing (small-collection-utilities.md).
Ingestion points: Data enters the system via strings processed by these utility functions (e.g., the text parameter in decodeEntities).
Boundary markers: The skill suggests using a standardized Resource envelope to wrap repository results at the domain boundary, facilitating clear separation between raw transport data and validated domain models.
Capability inventory: The provided logic is limited to string manipulation and arithmetic; it does not perform network operations, file system modifications, or subprocess execution on the processed data.
Sanitization: The documentation includes mandatory security warnings, stating that 'decoding is not sanitizing' and emphasizing that decoded output must be properly escaped before use in markup or database queries.

kmp-architecture-footguns