sdlc

Author / tune / validate / reverse-engineer / explain a canvas (a template + instructions pair under .claude/canvases/<name>/) via the canvas-author agent. Probing schema-aware dialog; never writes without confirmation. v1 ships tune + validate.

Generate the implementation strategy for a tracker issue and post it. In strict mode (default) sets state:awaiting-strategy-review and halts; in auto mode (per gate:auto label) sets state:strategy-approved directly and completes. Resolution by gate:* label → sdlc.yml.gate1_default.

Topology A — work one issue end-to-end with a flat team of split-pane teammates. Composes the team from sdlc.yml develop_team plus the issue's needs:* labels.

Topology B — parallel batch execution. Spawns one coordinator teammate per issue; each coordinator runs phase agents as headless subagents. Best for AFK throughput on pre-approved issues.

Decompose a request into a PR-sized issue plan. Produces a YAML at .ai-docs/plans/<slug>.yaml; iterates against your feedback until you approve. Does NOT touch the tracker.

Probing, schema-aware conversational authoring of canvases (template + instructions pairs at .claude/canvases/<name>/). Use when creating a new canvas, tuning an existing one, reverse-engineering a canvas from an example artifact, validating a canvas, or explaining how knobs affect produced artifacts. Never writes without confirmation; surfaces tradeoffs and constraint conflicts proactively. v1 ships tune + validate modes.

Read-only auditor that diffs the project's `claude-platform` skill against canonical Claude Code and Anthropic Agent SDK docs, surfacing renamed / added / removed fields, new lifecycle events, new SDK primitives, and changed precedence rules. Use when the user asks "is this still right", before authoring new components, or on a recurring cadence (`/loop`, `/schedule`). Does NOT edit files — reports only.

Per-issue coordinator for Topology B parallel orchestration. Spawned as a teammate by `/orchestrate`; spawns implementer/validator as **subagents** (Agent tool) — never as further teammates. SKELETON — not yet wired to a command.

Executes an approved spec. Refuses to start unless the issue carries `state:strategy-approved` (Gate 1 enforcement). Creates a branch, writes code per spec, opens a PR. Does not iterate strategy.

Decomposes an understood problem into PR-sized issues. Iterates a YAML plan across chat turns based on human feedback (Gate 0 — synchronous, in-chat). Halts only on explicit human approval; does not write to the tracker.

Meta-knowledge for authoring canvases — the template + instructions pairs that govern artifacts produced by SDLC agents (specs, plans, reports, …). Covers the canvas concept, the canonical knob taxonomy, conversation modes (new / tune / reverse / validate / explain), the response scaffold, and the anti-pattern catalog. Use when authoring a new canvas, tuning an existing one, reverse-modeling a canvas from an example artifact, validating a canvas, or teaching the canvas system to a new contributor.

Authoritative reference and authoring guide for the Claude platform — both the Claude Code `.claude/` config layer (skills, subagents, output styles, hooks, plugins, settings) and the Anthropic Agent SDK (Managed Agents API: agents, sessions, threads, multiagent, skills). Use when authoring a skill / subagent / output style / hook / plugin manifest, when designing a multi-agent system, or when comparing CLI vs SDK and translating between them.

End-of-session ceremony — articulate what's next, write `.ai-docs/handoff.md`, update memory, update CLAUDE.md if structural decisions landed, verify clean working tree. Symmetric counterpart to `/prime`. Use when wrapping up a session, before clearing context, or whenever the user signals they want to save state.

Load session context at cold-start — handoff, active Linear ticket (derived from branch), recent commits, stack position. Trigger when the user opens a fresh session or asks "where were we", "what's the state", "starting session", "/prime", or otherwise signals they want a status snapshot.

Workflow judgment for this project's SDLC loop — when to use /plan vs /design vs /develop vs /orchestrate, what each gate means, how to recover from a halt, how stacks and artifacts flow. Use when starting work on an issue, deciding the next command, interpreting a halt from any phase agent, or recovering from a failed gate.

Probing, schema-aware canvas authoring scaffold for developers. Activate when running `claude --agent canvas-author --output-style canvas-flow-developer` for marathon canvas-authoring sessions where the user is comfortable with system vocabulary (knobs, schema, sections) and wants every response in a rigid four-block decision-gated shape.

Conversational, outcome-framed canvas authoring scaffold for non-technical users. Activate when running `claude --agent canvas-author --output-style canvas-flow-seller` for users who think in outcomes (a daily brief, a follow-up email, a meeting summary) rather than in knobs and schemas. Hides the canvas system entirely by default; reveals more mechanism only as the user signals they want it.