early-prototype

Product / PM / Worker session-lifecycle suite — the timeteam skills, hooks, and agent that turn Claude Code sessions into auditable role-based work cycles.

12 skills · 1 agent · hooks

Agent-to-agent coordination over GitHub Discussions — agents open threads, join, reply, sign off and close, and read back who is in a thread and whether anyone replied. Includes a PR hook that opens a thread per pull request with a duplicate-claim check.

1 skill · 2 commands

Explicit session handovers: a subagent writes with bundled DrDoc guidance, asks the main agent for missing context, and returns the handover for review.

1 skill

Strips performed-quality tics from Claude's prose. Optional Dewormer output style plus a dewormer skill carrying the full ~90-term caution list across nine families: anthropomorphised code-speak, self-important spotlighting, performative candor, unsolicited validation, adverb inflation, the em-dash reframe, hedging connective tissue, metaphor soup, and sign-off tics.

1 skill

Enable, inspect, or disable persistent Project Knowledge Map context for a chosen Codex project. Requires an existing Knowledge Graph Kit checkout with its activation helper; Codex hook trust remains the owner's step.

1 skill

Claude Code MCP plugin for per-project markdown kanban boards, a REVIEW completion gate, and optional GitHub Projects sync. Separate manual MCP setup documented for Cursor and Codex.

1 MCP server

Any body of material as a dated reading note: answer first, sourced provenance, marked claims, numbers with scale and a baseline, a 2,000-word ceiling, and a closing decision section. Content agnostic, with a research special case. Ships the format, voice rules, template, worked example, a structural checker with H-number and EXP_ register checks, and an HTML page builder with a fixed white light default plus dark and auto themes.

1 skill