The Agent Skills Directory

[COMMAND_EXECUTION]: Persistence mechanism through system services. The skill provides commands such as pd install and scripts like preflight.sh to register and manage a background daemon via macOS launchd, allowing the agent to maintain a presence across sessions.
[COMMAND_EXECUTION]: Arbitrary shell command execution via the fleet engine. The pd-fleet.yml configuration (detailed in schemas/pd-fleet.schema.md) allows the definition of agents using a custom backend where the prompt field is executed directly as a shell command.
[REMOTE_CODE_EXECUTION]: Dynamic execution of tasks by background agents. The fleet mechanism facilitates spawning sub-agents (e.g., claude-cli) with specific tool access like Bash(npm test*), allowing automated code execution triggered by events such as git commits.
[CREDENTIALS_UNSAFE]: Local management of encryption keys. The daemon generates and manages an AES-256-GCM master key at ~/.port-daddy/master.key (referenced in SKILL.md and scripts/preflight.sh) used for encrypting and decrypting session notes.
[DATA_EXFILTRATION]: External data transmission capability. The skill supports an inconsistency_webhook for Slack (found in assets/.portdaddyrc.starter), which provides a channel for sending internal state or coordination data to an external service.
[PROMPT_INJECTION]: Indirect prompt injection attack surface. The coordination architecture (using notes and tuples) creates a shared memory space where data written by one agent can influence the subsequent behavior of other agents, such as the qa or cartographer agents, without explicit sanitization layers.

port-daddy