work-graph

The work graph format that plan-work creates and execute-work runs. Load it when you have to read, validate, hand-edit or archive a work graph (graph.md, nodes/<id>.md, history.md, backlog.md), map a

SKILL.md

GitHub

Work graph

A work graph is the resumable state of a program of work: graph.md holds the header, the node table and the Log; nodes/<id>.md holds one node's contract and plan; history.md holds archived rows and Log entries. The format is normative in references/graph-format.md; the templates references/graph-template.md and references/node-template.md start a new graph or node; references/ledgers.md maps the graph's operations onto another ledger when the operator uses one.

Use scripts/work_graph.py <command> <path to graph.md>:

  • check: validate the graph (cycles, unknown edges, contract fields, Consumes resolved to Produces, Files overlap without a dependency path, archive caps). Exit 1 with one line per defect.
  • ready: list the nodes whose dependencies are all done.
  • status: counts per state, the ready set, the schedule and the stuck nodes.
  • archive: move closed rows and old Log entries to history.md (--dry-run reports the moves). With --rows <json> the commands read a custom ledger's export instead of the table.

Planning a graph is the plan-work skill; running one is the execute-work skill. This skill only describes the files and the script they share.