The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses eval to process variables returned by the resolve_team_learn_paths function. This function parses the team_ai_directives path from a local configuration file (.adlc/init-options.json). If this file contains malicious shell metacharacters, it can lead to arbitrary command execution when the skill is initialized.
Evidence in scripts/team-learn.sh: eval "$(resolve_team_learn_paths)"
Evidence in scripts/helpers.sh: TEAM_AI_DIRECTIVES=$(grep '"team_ai_directives"' "${PROJECT_ROOT}/.adlc/init-options.json" | sed 's/.*"team_ai_directives"[[:space:]]*:[[:space:]]*"$[^"]*$".*/\1/' | head -1)
[DYNAMIC_EXECUTION]: The skill utilizes eval to dynamically define shell environment variables and employs awk to evaluate mathematical logic passed as strings.
Evidence in scripts/score-confidence.sh: The calc function executes input strings via awk's BEGIN block: awk "BEGIN { printf \"%.2f\", $1 }".
[INDIRECT_PROMPT_INJECTION]: The skill ingests and summarizes the current session history and git logs to extract 'Context Directive Records' (CDRs).
Ingestion points: Phase 1 in SKILL.md instructs the agent to review the current session and git history (git diff, git log) to identify reusable patterns.
Boundary markers: The instructions lack explicit boundary markers or directives to ignore instructions embedded within the session data being processed.
Capability inventory: The skill has the ability to write to the filesystem and perform git commit and git push operations on a remote repository.
Sanitization: The skill includes scripts/scrub-session.sh which uses sed to redact secrets (API keys, tokens) from summaries before they are recorded, and signal_gate in scripts/helpers.sh filters out low-value content.

team-learn