The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements a 'Multi-Goal Intake and Ledger Protocol' described in references/multi-goal-ledger.md. This protocol creates a persistent Markdown ledger file in ~/.agent/tmp/ that stores original user commands and normalized goals.
Ingestion points: The agent is instructed to reread this ledger from its absolute path at the start of every turn and display it in the session context.
Capability inventory: The system is designed to route tasks to specialized skills like nerd-execute and nerd-surgery, which likely possess file modification and command execution capabilities.
Boundary markers: The ledger uses Markdown structure and specific headers, but it does not employ robust sanitization or advanced delimiters to prevent embedded instructions within user-provided goals from influencing the agent's logic during the re-reading phase.
Sanitization: The protocol explicitly mentions redacting credentials and secrets from the ledger, which is a positive security practice, though it does not filter for prompt injection patterns.
[DYNAMIC_EXECUTION]: The skill includes a Python script scripts/prompt_hook.py designed to generate JSON-formatted prompt context for specific agent interfaces (claude-code, codex, cursor). This script is intended to be executed by the agent environment to modify the prompt dynamically.
[COMMAND_EXECUTION]: The SKILL.md file contains instructions for the user or developer to execute a local maintenance script: python3 scripts/validate_skills.py. This promotes the execution of local scripts to validate the skill family.
[PROMPT_INJECTION]: The scripts/prompt_hook.py script generates an INSTRUCTION string that mandates the agent to 'Always invoke the installed nerd-smart skill before substantive work on every user request', which is a behavioral override intended to force a specific workflow on the AI agent.

nerd-smart