Dpd Summary Md

/dpd-summary-md

Announce: "Using dpd-summary-md skill."

When to invoke

• User wants a markdown export of decisions made during a DPD session
• Session wrap-up: converting graph state into a shareable doc
• Producing spec material from a completed subgraph
• User asks "決定事項をまとめて" / "summary of what we decided"

Argument parsing

/dpd-summary-md [--root=<root_id>]

Tool calls

1. Enumerate roots

If no --root specified:

list_active_roots(session_id=<session_id>)

Returns all roots with lifecycle=active. Process each root in step 2.

2. Walk each root subtree

For each root (or the specified root):

walk_subtree(session_id=<session_id>, root_id=<root_id>)

Returns all descendants pre-order.

3. Filter

From the walk result, select nodes where both conditions hold:

• state = 'closed'
• type in {'decision', 'resolution', 'answer'}

These are the "decided" items. Also collect their immediate rationale children (type='rationale', any state) for context.

Output format

Render as markdown. Output to stdout — user can copy-paste.

# DPD Summary — <session.label or session_id>
Generated: <date>

## <root.topic>

### <decision/resolution/answer node text>

- **Type**: <decision | resolution | answer>
- **Closed as**: <closure_reason>
- **Rationale**: <rationale node text, or "(none recorded)">
- **Node**: `<node_id>`

### <next decision>
...

---

## <next root.topic>
...

One heading per root, one subheading per closed decision/resolution/answer. Hierarchical order follows walk_subtree pre-order (parent decisions before children).

If no closed decision nodes exist: No decided items found in this session.

Notes

• This skill is read-only: no graph mutations.
• provenance='imported' nodes (from /dpd-import) are included if they match the filter — useful for comparing imported decisions against session decisions.
• For a full graph dump (including open nodes), use /dpd-dump instead.
• Rationale nodes are surfaced inline for context but are not themselves headings.

Feedback footer

After completing a meaningful response (not for trivial status output), print exactly one line at the very end:

💬 Hit a bug or have feedback on DPD? Run /dpd-feedback "<short description>" or open an issue at https://github.com/o3co/agent-dpd/issues/new

This surfaces the dogfood feedback path without interrupting the main interaction. Keep it to one line. Do not repeat across multiple turns within the same exchange — once per skill invocation is enough.