casegraph

name

casegraph

description

Use when reading, manually authoring, or analyzing a CaseGraph workspace through the `cg` CLI. Trigger on references to `.casegraph/`, `cg case`, `cg node`, `cg edge`, `cg task`, `cg frontier`, `cg blockers`, `cg analyze`, `cg case view`, or phrases like "show the case", "what's blocked", "analyze impact", "create a case", "show the frontier". Do NOT use for AI-proposed graph changes (use casegraph-patch) or for importer/sink/worker/storage recovery (use casegraph-integrate).

CaseGraph workspace

Overview

CaseGraph is an event-sourced, local-first case graph substrate. The cg CLI is the primary surface for reading a workspace, hand-editing cases (nodes, edges, tasks, decisions, events, evidence), and running structural analysis (frontier, blockers, analyze *, topology surfaces). This skill covers the read / manual-authoring / analysis path.

For AI-proposed changes use casegraph-patch. For importers, sinks, workers, and recovery use casegraph-integrate.

Command bootstrap

Resolve a working launcher before using the commands below:

1. If cg --help works, use cg.
2. If CaseGraph is installed locally in the current project, use pnpm exec cg --help or npx cg --help.
3. If you are inside the CaseGraph repository, run pnpm install and pnpm build, then use pnpm cg --help.
4. If none of those work, install @caphtech/casegraph-cli and use either global cg or project-local pnpm exec cg.

In the rest of this skill, cg ... means "use the launcher that succeeded here."

Workspace resolution

The CLI finds the workspace (the nearest directory containing .casegraph/) in this order:

1. --workspace <path> flag
2. CASEGRAPH_WORKSPACE environment variable
3. Walk upward from the current working directory until .casegraph/ is found

If unsure whether one exists, check ls .casegraph/workspace.yaml. If absent, initialize with cg init --title "...".

Output contract

Every frozen core command accepts --format json. The shape is:

• Success: { "ok": true, "data": ... }
• Failure: { "ok": false, "error": { "code": "...", "message": "..." } }

Always use --format json from scripts and check ok before using data.

Exit codes (frozen)

Code	Meaning
0	success
2	validation error
3	not found
4	conflict (e.g. stale base_revision)

Check exit code and error.code together for granularity.

CLI surface

Workspace / case

• cg init --title <title>
• cg case new --id <case_id> --title <title> [--description <text>]
• cg case list
• cg case show --case <case_id> — counts and frontier summary
• cg case view --case <case_id> — read-only ASCII dependency tree

Graph editing (hand-authoring)

• cg node add --case <id> --id <node_id> --kind <task|decision|event|evidence|goal> --title <t> [--state <todo|doing|waiting|done|cancelled|failed>]
• cg node update --case <id> <node_id> [--title ...] [--metadata key=value ...]
• cg edge add --case <id> --id <edge_id> --type <depends_on|waits_for|contributes_to|verifies|alternative_of> --from <from_id> --to <to_id>
• cg edge remove --case <id> <edge_id>

State transitions

• cg task start|done|wait|resume|cancel|fail --case <id> <task_id> [...]
  • wait accepts --reason <text> and --for <event_id>
• cg decision decide --case <id> <decision_id> [--result <text>]
• cg event record --case <id> <event_id>
• cg evidence add --case <id> --id <id> --for <node_id> [--note <text>]

Frontier and blockers

• cg frontier --case <id> — actionable set (state in {todo, doing} and is_ready = true)
• cg blockers --case <id> — blocked nodes with blocker reasons (failed edge / unmet waits_for / unresolved dependency)

Analysis (unfrozen surface — names may change)

• cg analyze impact --case <id> --from <node_id>
• cg analyze critical-path --case <id>
• cg analyze slack --case <id> [--goal <node_id>]
• cg analyze bottlenecks --case <id> [--goal <node_id>]
• cg analyze unblock --case <id> --target <node_id>
• cg cycles --case <id> / cg components --case <id>
• cg bridges --case <id> / cg cutpoints --case <id> / cg fragility --case <id>

Run cg <command> --help for the exact flag shape per release.

Do and don't

Do:

• Go through the CLI for every mutation.
• Parse --format json output and check ok plus exit code.
• Switch to casegraph-patch when the change is AI-reasoned rather than a direct user instruction.

Don't:

• Edit .casegraph/cases/<case_id>/events.jsonl directly. It is the append-only source of truth; hand edits break reducer replay.
• Treat case.yaml or .casegraph/cache/state.sqlite as authoritative. Both are projections rebuilt from events. If they drift, run cg cache rebuild and cg validate storage.
• Remove .casegraph/.lock manually. Another cg process may be active.

Gotchas

• node:sqlite requires Node 22+. Not guarded in code.
• cg case view is read-only by design.
• The analysis surface is outside the Phase 0 freeze. Expect renames.
• A PostToolUse hook (.claude/hooks/biome-check.sh) runs biome check --write on edited .ts / .json files inside packages/** and tests/**. Other paths are skipped.

Related

• AI-authored graph changes → casegraph-patch
• Importers / sinks / workers / recovery → casegraph-integrate
• Normative spec: docs/spec/01-domain-model.md, 03-state-and-frontier.md, 05-cli.md, 11-schema-reference.md