stakeholder-map

name	stakeholder-map
description	Use when the user says /stakeholder-map, asks to build a stakeholder/political-topology map for a new leadership role, wants to review coverage of their relationship-building, or asks for meet-in-what-order guidance during onboarding.
disable-model-invocation	true

Stakeholder Map

Two modes sharing one memory-graph backbone:

• leader-onboarding — bootstrap the stakeholder map (manual form + optional calendar seed), tag people with power/category/function, render a stakeholder chart.
• coverage-review — periodic audit over the populated graph that surfaces gaps, echo-chamber signals from advice content, and a recommended-next-interviews list.

Announce: "I'm using the stakeholder-map skill to [build your stakeholder map / review your coverage]."

Flow: Prerequisites → Invocation → Mode Detection → Mode Dispatch.

Reference files (read on demand, not upfront):

• graph-schema.md — tags, relations, write protocols, shared computation rules
• bootstrap.md — manual form + calendar-seed flow (Mode A intake)
• chart-render.md — Mode A chart rendering
• heatmap-render.md — Mode B heatmap rendering
• coverage-queries.md — five dimension queries, echo-chamber, structural-gap
• sync.md — --sync drain line format and partial-replay semantics

Privacy

Person entities contain sensitive assessments. Storage is local memory MCP only — never export to repo or share the raw graph.

Prerequisites

Verify memory MCP: mcp__memory__read_graph. If unavailable, warn and set a session-wide flag so ALL writes route to pending-sync/ for the remainder (not just the current write). The file format mirrors 1on1-prep — see ../1on1-prep/pending-sync.md. Check for existing pending-sync files on entry.

Rendering requires the excalidraw canvas to be reachable via Preview. Check at render time (not skill invocation) using mcp__Claude_Preview__preview_list and mcp__excalidraw__describe_scene. On failure, fall back to text output and print:

Excalidraw canvas not reachable. To enable visual output, run preview_start("excalidraw-canvas") and re-render with /stakeholder-map --render.

Invocation

/stakeholder-map --mode=leader-onboarding [--seed-from-calendar [--days=30]]
/stakeholder-map --mode=coverage-review
/stakeholder-map --render                     # re-render last output
/stakeholder-map --sync                       # drain pending-sync

--sync drains pending-sync files in skills/stakeholder-map/pending-sync/ into the memory graph. Line format, partial-replay semantics, and authoring rules are specified in sync.md.

Mode Detection

If no --mode flag:

1. mcp__memory__read_graph — count entities that satisfy the stakeholder-map membership predicate defined in graph-schema.md.
2. If count is 0 → default to leader-onboarding and confirm with user.
3. If count < 10 → default to leader-onboarding (resume mode) and confirm.
4. If count ≥ 10 → default to coverage-review and confirm.

Always confirm the auto-detected mode before running. The user can override.

Mode Dispatch

leader-onboarding

Load bootstrap.md and run the flow. On completion (or every 5 stakeholders added), load chart-render.md and render.

coverage-review

Preconditions:

• Graph has ≥10 entities tagged with any stakeholder-map tag.
• If below threshold: "Graph is sparse ({count}). Coverage review works best with ≥10. Continue anyway or run leader-onboarding first?"
• If empty: exit with "Graph is empty. Run leader-onboarding first."

Load coverage-queries.md and run the queries. Load heatmap-render.md and render.

Common Mistakes

• Minting enum values at runtime. Only the canonical values in graph-schema.md are writable. If the user says "he's a distinguished engineer", don't invent [role:distinguished] — record the title in a [context] observation and tag [role:staff] (closest canonical).
• Skipping the replaceable-tag write protocol. Writing a new [power:formal:*] without first deleting the prior one leaves two values on the entity and breaks latest-tag selection.
• Creating a relation before both entities exist. create_relations requires both ends to resolve. In bootstrap, create the target entity first, or skip the relation.
• Using stakeholder-map-tagged entities as a general contact database. [context]-only entities (from seen-but-skip or 1on1-prep upgrades) are NOT on the map per the membership predicate. Don't let them leak into charts or coverage counts.
• Rendering without preflight. Always run preview_list + describe_scene before drawing. If either fails, use the text fallback — do not silently skip the render.

Compatibility With 1on1-prep

Both skills write to the same memory graph using the same observation format. Stakeholder-map's tag schema is documented in graph-schema.md and is cross-referenced from 1on1-prep's schema. Running either skill first is valid; stakeholder-map's bootstrap offers an upgrade path for entities created by 1on1-prep.

Where this skill persists state

Per ADR #0020, per-leaf data class assignment against the six-leaf decision tree:

memory MCP knowledge graph (mcp__memory__* from @modelcontextprotocol/server-memory):

• Stakeholder graph: person/team entities, observations, relations, power / category / function tags. Confidentiality contract per the Privacy section above: never export raw graph contents to repo or share verbatim — honors the local-only-sensitive contract ADR #0020 names for this destination.

Not used by this skill: auto-memory MD, ruflo MCP, user working repo (no committed exports — Privacy forbids them), scheduled-tasks MCP, plugin-internal memory (decisions.md / patterns.md).

Out-of-scope local files (not memory layers): pending-sync/ (transient buffer when MCP unavailable, drained by --sync), excalidraw canvas (rendering sink, not state).