The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill functions as an orchestrator that ingests data from external sources, specifically Slack issue reports and GitHub pull request comments. This content is processed by the agent to perform triage, reproduction, and babysitting tasks.
Ingestion points: Slack thread content is read in adapters/cursor/automations/benny/skills/triage-issue-reports/procedure.md and pull request comments are queried in playbooks/babysit.md.
Boundary markers: The skill lacks explicit structural delimiters when interpolating untrusted data, but it consistently instructs the agent to treat such data with skepticism.
Capability inventory: The orchestration scripts in scripts/ have the capability to execute shell commands (git, gh) and spawn child agent processes using different model providers.
Sanitization: The references/bugbot-triage.md file provides a detailed procedure for skeptical evaluation of bot and human comments.
[EXTERNAL_DOWNLOADS]:
The scripts/bootstrap.ts utility programmatically triggers bun install to fetch and update Node.js dependencies defined in scripts/package.json from the official registry.
The adapters/cursor/make-bot-ui.md documentation provides instructions to download the Tailscale installer from https://tailscale.com/install.sh.
[COMMAND_EXECUTION]: The skill frequently executes system-level commands to perform its core engineering tasks. This includes using git for source control, the gh (GitHub) and origin (internal) CLIs for pull request management, and launching secondary agent processes via claude, codex, and grok command-line tools.
[PRIVILEGE_ESCALATION]: In the adapters/cursor/make-bot-ui.md guide, the agent or user is instructed to use sudo to install Tailscale and bring up its networking node. This use of elevated privileges is restricted to the setup of a well-known service.

pstack