orchestrator-status

Returns a structured snapshot of the SDD Orchestrator's current state: whether intent classification is enabled, active SDD changes in progress, and skills registered in CLAUDE.md.

Triggers: User invokes /orchestrator-status

Process

Step 1 — Read CLAUDE.md

Read the project's CLAUDE.md (located at the project root, e.g., C:/Users/juanp/claude-config/CLAUDE.md or the current working directory's CLAUDE.md).

Extract:

• Whether intent_classification: disabled appears under any ## Always-On Orchestrator — Override section. If absent, classification_enabled = true.
• Count the numbered items under ## Unbreakable Rules (each ### N. or numbered list item = 1 rule).
• The file path used as configuration_source.

Step 2 — Scan active SDD changes

List directories under openspec/changes/ (non-recursively). Exclude archive/.

For each non-archived directory:

• Record name (directory basename)
• Identify status:
  • If verify-report.md present → "verified"
  • Else if tasks.md present → "in-progress"
  • Else if design.md present → "designing"
  • Else if proposal.md present → "proposed"
  • Else → "unknown"
• List artifact filenames present (proposal.md, exploration.md, design.md, specs/, tasks.md, verify-report.md)

Step 3 — Extract loaded skills

From CLAUDE.md's ## Skills Registry section, collect all lines matching `~/.claude/skills/<name>/SKILL.md` pattern. Extract <name> values.

Count total as skills_registry_count.

Identify orchestrator + SDD phase skills specifically:

• Core: sdd-ff, sdd-new, sdd-status, orchestrator-status
• Phases: sdd-explore, sdd-propose, sdd-spec, sdd-design, sdd-tasks, sdd-apply, sdd-verify, sdd-archive

Step 4 — Build JSON output

Emit a fenced JSON code block:

{
  "orchestrator_state": {
    "classification_enabled": <true|false>,
    "unbreakable_rules_count": <N>,
    "session_start": "<ISO 8601 timestamp>",
    "configuration_source": "<absolute path to CLAUDE.md>"
  },
  "active_sdd_changes": [
    {
      "name": "<change-dir-name>",
      "status": "<proposed|designing|in-progress|verified|unknown>",
      "artifacts": ["<filename>", ...]
    }
  ],
  "loaded_orchestrator_skills": [
    "<skill-name>",
    ...
  ],
  "skills_registry_count": <N>
}

Step 5 — Write prose interpretation

After the JSON block, output a prose section titled ## Interpretation:

Orchestrator Status (<date> UTC)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Orchestrator: ENABLED  (or DISABLED if classification_enabled is false)
  Rules loaded: <N> unbreakable rules from CLAUDE.md
  Configuration: <configuration_source>

Active SDD Changes: <count>
  (list each: • <name> (<status>) — Artifacts: <comma-separated artifact list>)
  (if zero: "None — no active changes found in openspec/changes/")

Loaded Skills: <orchestrator-skill-count> orchestrator + SDD phase skills
  Core: <list core skill names>
  Phases: <list phase skill names>
  Project catalog: <skills_registry_count> total skills

Ready to accept: /sdd-ff <slug> | /sdd-explore <topic> | /sdd-new <change>

Rules

1. Read-only: This skill MUST NOT modify any files. It is a pure read/inspect operation.
2. No external calls: Do not make network requests or invoke shell commands that mutate state.
3. Classification check: Always check for the Override section in CLAUDE.md before defaulting classification_enabled to true.
4. Exclude archive/: Never count archived changes as active. Only list entries directly under openspec/changes/ that are NOT under openspec/changes/archive/.
5. Format contract: Output MUST contain the JSON code block first, then the ## Interpretation prose section. Do not reverse the order.
6. Graceful fallback: If openspec/changes/ does not exist, set active_sdd_changes: []. If ## Skills Registry section is absent from CLAUDE.md, set skills_registry_count: 0 and loaded_orchestrator_skills: [].