Operating model

This is the reference detail behind the the-loop skill — the operating model an agent harness follows once a work item's phase is underway. Each page below is read by the harness at the point in the loop where it's relevant (progressive disclosure — see token economy); read them here for the same detail, up front.

Reference	Covers
Workflow	The loop, phases, TDD, reviews, autonomy, DAG, resumability.
Context	Context-window management: clearing vs. compaction, the checkpoint-then-reset protocol, per-harness mechanics.
Onboarding	The guided, schema-driven config onboarding /the-loop:init runs.
Instructions	User-provided custom instruction docs: when to read them, precedence, what they can and cannot override.
Design artifacts	UI/UX design artifacts (Figma / HTML prototypes) in the design phase and the designer iteration loop.
Reviewing	The self/critic review procedure the review counts drive.
Security	The security lens on every phase gate: threat-model-lite, security design, the security-review gate, human sign-off tiers.
Tooling	Repo management, per-language tooling matrix, hooks, CI parity.
Testing	Gherkin scenario docstrings on integration tests, the queryable scenario view, OpenAPI/GraphQL contract conventions.
Minimalism	The generation-time decision ladder to counter code bloat.
Token economy	Token/cost levers (model routing, verbosity, disclosure, sub-agents, telemetry); advisory, never at the expense of rigor.
Collaboration	Personas/roles, paper trail, conflict log, messaging, MCP.
Observability	Dev==runtime logging, levels, browser logging.
Automation	Distribution, the CLI, webhooks, predictability, learnings lifecycle.

The source SKILL.md that ties these together lives at skills/the-loop/SKILL.md in the repository.