/start — Session Resume Skill

Invocation: /start (also called automatically from agent activation files) Works for: any agent (saga, freya, mimir, idun)

Purpose

Detects whether a previous session was saved for the active agent and offers to resume it. If no state file exists, proceeds silently with the normal activation sequence.

Behavior When Invoked

1. Detect Active Agent

Identify which agent is currently active. Look for _bmad/_state/[agent].md in the current project repo.

2. Load State

Primary: local file

Check for _bmad/_state/[agent].md in the project root. This is the authoritative source.

Optional: Agent Space enhancement

If Agent Space is configured, also call session-start — but only as an enhancement, never a requirement:

curl -X POST "{BASE_URL}/functions/v1/session-start" \
  -H "Authorization: Bearer {API_KEY}" \
  -H "Content-Type: application/json" \
  -d '{
    "agent_id": "[agent]",
    "project": "[repo-folder-name]",
    "model_target": "claude"
  }'

If the call succeeds and presence.last_status_report is present, use it as the state source only if it is more recent than the local file (e.g., previous session was on a different machine). If the call fails for any reason, continue silently with the local file.

The response may also contain:

session_id — register as active session ID if present
instructions — skill chain overrides. Load any levels present.
files — cached design-process folder. Display file count if any.
messages — unread messages. Show if any.

Fallback chain: local _bmad/_state/[agent].md → Agent Space presence → fresh start

3. If State Found

Display the previous session summary clearly:

⏸ Previous session found ([date from Wrapped field])

Where I left off: [content from "Where I Left Off" section]
Next action: [content from "Next Action" section]

Resume where we left off, or start fresh?

Wait for the user's response.

If resume:

Read the full state file, including the Context and Open Questions sections
Jump straight to the Next Action — no scanning, no re-introduction, no status report
Treat the context as already established — don't re-explain what was already known

If fresh:

Proceed with the normal activation sequence for this agent
Do not delete the state file (the user may want to refer back to it)

4. If Nothing Found

No local file, no Agent Space record — proceed with the normal activation sequence. Do not mention /start or the absence of a state file.

Notes

The state file is written by /wrap. If no /wrap was run at the end of the previous session, there will be no file to find.
The state file lives at _bmad/_state/[agent].md relative to the project root.
Agent Space is optional — local file works without it.
On resume, prioritize getting back to work quickly. The user already knows the context — they don't need a recap beyond what's shown in the summary.

3.0 KiB Raw Blame History