CommunityCoding & Developmentgithub.com

a-freshcat/workspace-main-conversation

Manage conversation division of labor, file ownership, and claim consolidation in a shared workspace. Use when entering an active workspace, performing shared writes, or organizing scattered artifacts.

What is workspace-main-conversation?

This skill enables structured collaboration within a shared workspace by managing conversation roles, file ownership, and task tracking with minimal state overhead. It is designed for AI agent platforms like Codex and DSH. The skill distinguishes between 'main' and 'child' conversation roles: the main conversation aggregates tasks, reviews alerts, and checks for conflicts and duplicate work, while child conversations only maintain their own task records and artifacts. File access and writes are serialized through a registration system to prevent conflicts, and a summary file (`sum.md`) is generated by the main conversation script. When information is incomplete, the skill can fetch additional details using pagination (`more=true`) or perform detailed file checks with `check --offset N`. It also provides guidance for handling scattered artifacts, claims, and handoffs, ensuring coordination without replacing tool permissions or blocking uncoordinated external writes.

Works with~Claude Code✓Codex CLI~Cursor
npx skills add a-freshcat/workspace-main-conversation

Installed? Explore more Coding & Development skills: steipete/bluebubbles, steipete/eightctl, steipete/blucli · View all 6 →

Ask in your favorite AI

Open a new chat with this agent skill pre-loaded.

Documentation

Workspace Collaboration v2.1

Goal: Support collaboration with minimal state; read details on demand. Specific task assignments are at the user's discretion. Script paths are relative to this skill directory; workspace parameters use absolute paths. Compatible with Codex and DSH.

Entering or Resuming

Run node scripts/workspace.mjs context --workspace WORKSPACE --session SESSION. SESSION uses the current conversation ID provided by the runtime environment; if unavailable, keep as unknown—do not guess identity or claim. If the workspace is not enabled, continue with the user's original task; only execute init when the user explicitly designates a main conversation and enables management. For a new context, fetch the full brief first; for subsequent contexts in the same conversation, use the last fully processed --since STAMP and read only changes. scope=declared-metadata only checks registered information; it does not prove that source files have not changed or that unregistered conversations do not exist.

Execution

  • role=main: Aggregate registered tasks, process alerts, review conflicts and duplicate work; at phase end, generate sum.md using summary.
  • role=child: Only maintain own task records and artifacts; when context is needed, prefer detail --task ID.
  • When taking a task, changing status, or handing off results, use task --file INPUT.json to register; see the Command Reference for fields and commands. Normal replies do not require ledger updates.
  • Before shared writes, register writes first; if a claim conflict arises, serialize the handoff. Use check to verify registered paths, and re-read the target file before editing.
  • sum.md is generated by the main conversation script; child conversations update their own records. If content is unchanged, do not increase the task version.

When Information Is Insufficient

more=true uses the same --since and --offset NEXT to read all relevant pages, then saves the stamp. Unknown states, incomplete scans, or parse errors are retained as issues; use check --offset N for detailed file inspection. When scattered artifacts, unclear ownership, or unclaimed claims are found, read Claims and Consolidation. Only read Forensic Commands when DSH logs are needed as clues; for enabling old workspaces, see Migration. In daily use, only read the brief. Depend only on currently available state; use an explicit handoff to take over a main conversation. File timestamps do not indicate whether a conversation is running. These records coordinate cooperation, do not replace tool permissions, and do not prevent external writes that do not follow the conventions.

Related Skills