Human-readable guidance lives here when it does not need fixtures, schemas, or runtime binding.
session-evidence/: bounded session collection, Task Episode diagnostics, reader-facing session insights, and usage-efficiency evidence.project-harness/: static project evidence, core-change inspection, observability, design contracts, acceptance controls, and recovery safeguards.agent-customize/: agent instructions, Skills, MCP, Memory, Hooks, Custom Agents, Plugins, and platform-specific asset guidance. Start withagent-customize/routing.md.bootstrap/: the 0 -> 1 move for a project with no coding-agent harness yet: the specification structure contract and stack-specific spec examples. Load this when a requirement is not yet complete enough to implement against.loop-engineering/: repeated-work and schedule-ready owner selection. Load this when a task may become a Skill, automation, hook, command, script, custom agent, MCP-backed loop, or rule.tool-runtimes/: support contracts for runtime discovery and execution boundaries. These are not practice domains.
Detector and signal guidance stays with its owning model, executable capability, or skill-local contract. Promote shared prose only when two visible workflow consumers need the same owner; do not recreate a generic detector bucket.
references/coding-agent-practices/was split by evidence owner. Usesession-evidence/,project-harness/, oragent-customize/.references/reliable-delivery/,references/ai-friendly/,references/coding-agent-observability/, andreferences/ai-friendly-engineering/were folded intoproject-harness/.references/coding-agent-practises/was the historical misspelled path. Do not recreate it; route the document to its current evidence owner.references/harness-practises/was the historical misspelled harness path. Do not recreate it; usereferences/project-harness/.references/harness-engineering/was the former broad project domain. Do not recreate it; usereferences/project-harness/.