# derived: derive-agent-pack.mjs # Labs-FinTecAI Agent Pocket Guide Curvature Labs FinTec AI Ecosystem knowledge and governance repo. ## Entry sequence 1. Read `for-agents/llms.txt` (this file — pocket guide) 2. Read `for-agents/manifest.json` (structured routing with pack mode) 3. Read `for-agents/machine-context.json` when ecosystem object boundaries, interfaces, or source coverage matter 4. By task type, read the corresponding pack or source: - Project onboarding → `for-agents/projects//` - Topic maintenance → `for-agents/topics//` (derived from agent-pack.yaml) - Ecosystem-wide → `for-agents/ecosystem/` - Change proposal → `governance/change-protocol.md` (read source directly) - Controller recovery → engineering repo or `execution//state.md` ## Mode semantics - `source_direct`: read the agent entry, then consume mapped canonical sources with source refs. - `topic_manifest`: read the topic manifest source and generated topic pack. - `derived`: consume generated summaries/specs/anchors and verify coverage metadata. ## Hard rules - `for-agents/**` is derived. Do not hand-edit generated files. - Change fact-of-record Markdown through `governance/change-protocol.md`. - Terminology aligns to `ecosystem/glossary.md`. - Preserve source refs in every conclusion. - Consume `for-agents/machine-context.json` before project context files when object boundaries or interfaces matter. - Use `mode` in manifest.json to decide whether to consume an agent entry or canonical source directly. - For topic work, consume `for-agents/topics/index.json` before a topic-specific manifest. - Keep execution artifacts in the engineering repo; this repo carries knowledge, governance, and source mapping.