The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides instructions to construct shell commands by interpolating user-provided questions and environment variables directly into strings without specifying escaping or sanitization requirements. This presents a risk of command injection if the executing agent does not provide its own safety layer for shell calls.
Evidence:
In SKILL.md: obsidian-wiki graph-query "$OBSIDIAN_VAULT_PATH" "<question>" --pretty takes input directly from the vault path variable and the user's question.
In SKILL.md: ${QMD_CLI:-qmd} query $'lex: <key terms>\nvec: <question rephrased as a description>' interpolates user-controlled semantic terms into a CLI search command.
In SKILL.md: Grep -A 10 -B 2 "<query-term>" <candidate-file> uses user-supplied query terms in a grep command line.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and synthesize data from an Obsidian wiki, which constitutes untrusted external input. Malicious instructions embedded in wiki pages could influence the agent's output or synthesis.
Ingestion points: The skill reads page content from various wiki directories including concepts/, entities/, synthesis/, and projects/, as well as index.md and hot.md.
Boundary markers: While the instructions specify an answer structure ("Based on the wiki:"), there are no explicit boundary markers or "ignore embedded instructions" delimiters used when the agent processes the retrieved text.
Capability inventory: The skill possesses the capability to execute shell commands (obsidian-wiki, qmd, grep), perform local file reads, and append to a log.md file.
Sanitization: The skill does not define any validation, filtering, or sanitization protocols for the wiki content before it is used for synthesis.

wiki-query