delivery

Plugin delivery from marketplace devagent in mrjones-cs/devagent. Commands read from the repo's manifest — 23 days ago.

Skills (8)

• brainstorm

  Interactive requirements discovery for one feature - goals, users, data handled, client constraints and acceptance criteria. Proposes the S/M/L tier from explicit triggers, records the client AI-use position, and writes 01-requirements.md. Never designs.

• design

  Writes a feature's 02-design.md from its requirements - architecture, data classification, threat model, cost, observability, the personal_data, calls_model and has_ui flags, and load-bearing assumptions to verify - applying each standard the design touches.

• load

  Full session briefing for this engagement workspace - the NOTES.md current section, the docs/knowledge index, every active plan's progress and next ready task, build lock state and repository state, all labelled as data. Read-only.

• plan

  Runs the planning chain for a feature - design, research, spec-panel, workflow - in order, resuming at the first missing plan file, pausing only for decisions flagged for the operator, and ending with a summary.

• research

  Verifies a feature's load-bearing design assumptions - public-facts register first, then web-search leads confirmed only on allowlisted official pages - and writes 03-research.md with the engine and every query sent, applying corrections to 02.

• save

  Closes a session - updates NOTES.md, writes project lessons to docs/knowledge, sets the newest plan file's status, and reports uncommitted work, orphaned worktrees, locks and plan drift. With --handover exports the leak-reviewed deliverable; with --closeout inventories every copy.

• spec-panel

  Reviews a feature's 02-design.md through independent lenses, one fresh reviewer per lens in parallel, at most five findings each; merges and dedupes them, routes decisions to the operator, applies fixes to 02, and writes 04-review.md.

• workflow

  Writes a feature's 05-implementation.md from the final design - small tasks with files, acceptance, a single allowlisted verify command and a timeout - refusing while a decision is open or a load-bearing assumption is unverified.

Subagents (4)

• builder

  Implements exactly one task from 05-implementation.md in its own git worktree and leaves one WIP commit on the task branch. Dispatched by /delivery:build only; build verifies and integrates.

• requirements-analyst

  Drafts a feature's 01-requirements.md from the brainstorm conversation. Dispatched by /delivery:brainstorm only, with the answers already gathered; records an answer or an explicit `none stated` for every question an enterprise client's security and architecture review asks before design starts.

• researcher

  Verifies a batch of assumption rows handed to it by /delivery:research. Leads from the web-search script, confirms only on allowlisted official pages, and returns 03-research rows plus every query sent. Never reads 01 or 02 whole.

• reviewer

  Reviews a design through exactly one lens whose text /delivery:spec-panel inlines into the prompt. Read-only; returns at most 5 findings in the 04-review grammar. Always a fresh dispatch, never a fork.

Hooks (2)

• SessionStart

  python3 "${CLAUDE_PLUGIN_ROOT}/scripts/briefing.py"

• PostToolUse

  python3 "${CLAUDE_PLUGIN_ROOT}/scripts/briefing.py" --post

Similar plugins