WorkMesh is a docs-first project and task orchestration system for developers and coding agents.
It keeps planning state next to the code:
- tasks in
tasks/ - repo-local context in
.workmesh/context.json - durable decisions in
.workmesh/truth/ - global continuity in
~/.workmesh/
By default, actionable tasks must include Description, Acceptance Criteria, and Definition of Done, and Definition of Done must be outcome-based. Those task-quality requirements are configurable in .workmesh.toml or global config when a repo wants a looser or stricter policy. Incomplete work should stay in Draft or Needs Refinement until it is ready for To Do or In Progress.
Maintenance fixers can backfill missing task UIDs, remove broken dependencies, repair duplicate task IDs, and normalize drifted task filenames from task metadata.
Verify installed binaries:
workmesh --version
workmesh-mcp --versionBuild from source:
git clone git@github.com:luislobo/workmesh.git
cd workmesh
cargo build -p workmesh
cargo build -p workmesh-mcpCodex MCP example:
[mcp_servers.workmesh]
command = "/usr/local/bin/workmesh-mcp"
args = []Canonical WorkMesh skills live under skills/.
Installed agent paths follow the shared Agent Skills layout:
- project scope:
.agents/skills/for Codex and Cursor,.claude/skills/for Claude - user scope:
~/.codex/skills/,~/.cursor/skills/, and~/.claude/skills/
Each skill is self-contained and includes its referenced doctrine files inside its own skill root.
Primary documentation:
Supporting references:
Agent-readable mirror: