# LiFoli Works Agent Context LiFoli Works is not an ATS or single-domain HR SaaS. HR recruiting operations are the first beta domain pack for proving a general workflow execution engine. Core thesis: ```txt work capture -> service and intent detection -> connector/capability generation -> skill or recipe candidate -> approval-gated execution -> evidence, artifacts, audit logs -> memory and skill improvement ``` Agent Experience posture: - Agent-readable docs are first-class product surfaces. - Token-efficient discovery is available at `/.well-known/ai`. - Agent skills are packaged with `SKILL.md`, manifests, examples, and evals. - External skills from catalogs such as VoltAgent `awesome-agent-skills` are reviewed references, not automatically trusted runtime code. - The permission ladder is read, draft, local_execute, external_write, payment. - Human and agent actions must be distinguishable in audit records. - Agent-facing analytics preserve selected route, actor, approval class, and evidence state without exposing secrets. - Scoped credential lifecycle support should include create, rotate, revoke, and expire semantics before autonomous external writes are broadened. Automation route: ```txt Official MCP -> Browser automation -> PrintingPress-generated MCP -> human-in-the-loop fallback ``` Important documents: - `/docs/agent-experience.md`: public AX posture, route order, permissions, audit. - `/docs/api-recipes.md`: public examples for `/ax/scan`, `/mcp`, and skill discovery. - `/mcp`: runtime stdio MCP export manifest for read-safe LiFoli capabilities. - `/HEARTBEAT.md`: autonomous and scheduled agent check-in contract. - `/robots.txt`: legitimate assistant access policy and discovery pointers. AX Scanner: - Scores explicit readiness signals before any live crawling. - Selects official MCP, API connector, browser automation, PrintingPress-generated MCP, or human fallback. - Assigns a minimum permission class for the selected route. Current seed skill: - `recruiting_screening_brief`: create a recruiting screening brief from applicant/job inputs with evidence, redaction, and approval gates.