Start with the job. Follow the route.

Install Guild Stack on a receipt-verified host or preview one of 16 named targets, then run one bounded task with local .guild/ evidence.

Follow one task through Guild Stack's six-phase lifecycle, three entry lenses, approval gates, specialist lanes, review, and reflection.

The flat /guild:<verb> command surface: run the lifecycle, jump phases, inspect status, query memory, tune settings, and keep approval gates visible.

Install Guild Stack through the supported host path, choose stable or beta, inspect the install receipt, and apply updates without hiding host limits.

How Guild Stack scopes work across workspace, project, optional initiative, and run — and where durable state belongs at each level.

Meet Guild Stack's roster: 15 domain specialist templates minted on demand, plus the advisor and developer machinery agents, for scoped lanes and typed handoffs.

Guild Stack gives each specialist a focused context bundle before dispatch: task lane, required skills, relevant wiki pages, and no project-wide dump.

How Guild Stack turns an approved plan into specialist tasks, selects an execution mode, records handoffs, and recovers failed lanes.

How Guild Stack challenges AI work before trusting it: independent review packets, bounded rounds, recorded verdicts, and no silent weak sign-offs.

The migration-specific evidence packet, reviewer requirements, and followup policy used to verify work against the host-adapter migration spec.

Guild Stack settings: one settings.json file, seven-source inheritance, closed-key validation, model tiers, dispatch mode, security, and MCP pins.

Guild Stack routes each specialist lane to the cheapest model tier that fits the work, with deterministic scoring, overrides, and advisor escalation.

Use /guild:learn to map a codebase, build a knowledge graph, produce an onboarding tour, analyze changes, and approve deeper scans before they run.

Guild Stack records canonical project memory as gated Markdown under .guild/wiki/, then derives recall indexes, graph views, and specialist context from it.

Guild Stack can learn from real runs without silent mutation: reflections propose better skills or specialists, evals test them, and you approve promotion.

Use initiatives when a Guild Stack goal spans sessions, branches, or repos: attach runs by choice, track evidence, and close only after exec, release, and docs-sync gates.

Use Guild Stack across a repo of repos without copying knowledge up: workspace manifests, child wikis, read-through recall, and settings inheritance.

Run release checks and operations without losing control: computed quality verdicts, approved runbooks, dry-runs, and hard safety rails.

What Guild Stack may touch, what always requires approval, and how repo content, secrets, network access, and MCP tools stay inside explicit boundaries.

Inspect what Guild Stack did after a run: manifests, provenance, handoff receipts, verification, event logs, and local dashboard views under .guild/runs/.

Launch the benchmark dashboard on 127.0.0.1, import recorded Guild Stack runs, and inspect timelines, knowledge, and wiki evidence without a cloud service.

Add an opt-in Claude Code status line for active Guild Stack runs: loop round, cap, loop mode, restart count, and phase signal when available.

Guild Stack internals: one lifecycle, module-owned resources, host adapters, file receipts, settings resolution, and durable .guild/ evidence.

The Guild handoff bus: specialists write typed receipts under .guild/runs/, and leads, review, verification, and replay consume files.

The canonical Guild Stack vocabulary for nesting, lifecycle, host, substrate, review, knowledge, and model tiering.

Move from Guild Stack v1 to v2: command renames, settings.json migration, rigor flags, phase verbs, and the approval-gated lifecycle.