swot

name	swot
description	Use when the user says /swot, "landscape analysis", "SWOT analysis", "strengths and weaknesses", or wants to capture, review, or challenge organizational observations.
disable-model-invocation	true

SWOT Landscape Analysis

Structured SWOT for onboarding. Accumulates observations across sessions via the knowledge graph. Supports conversational capture, artifact-pointed capture, challenge, review, and multi-format export.

Announce: "I'm using the swot skill to help you build your landscape analysis."

Reference files (read on demand, not upfront):

• graph-schema.md — entity format, tags, provenance, examples
• pending-sync.md — file format, sync flow, error handling
• capture-form.md — observation-first capture form, parsing, auto-tag confirm flow
• challenge-checks.md — 4 quality checks with examples
• export-formats.md — markdown, excalidraw, presentation export

Prerequisites

Verify memory MCP: mcp__memory__read_graph. If unavailable, warn and route writes to pending-sync. Check for existing pending-sync files.

Invocation

/swot <org-name> [--mode=add|review|challenge] [--read <path-or-url>] [--from=1on1-prep] [--sync]

--sync drains pending-sync files (see pending-sync.md). --read triggers artifact-pointed capture. --from=1on1-prep ["names"] drafts SWOT observations from captured 1:1 discovery notes; optional comma-separated person names narrow the source (see "Discovery Handoff").

Confidentiality Refusal

When the caller passes --read <path> AND the path is local (not a URL), MUST run the refusal guard before reading the file:

bun run "$CLAUDE_PROJECT_DIR/skills/onboard/scripts/onboard-guard.ts" refuse-raw <path>

The guard is a no-op for URLs and paths outside any /onboard workspace — exits 0, /swot proceeds normally.

Exit codes, repo-root resolution, and override policy: see ../onboard/refusal-contract.md.

Org Lookup

mcp__memory__search_nodes({ query: "<org-name>" }) — exact match (with " SWOT" suffix) → use it, one substring → use it, multiple → ask, not found → Bootstrap, tool error → stop.

Bootstrap (New Org)

Requires memory MCP. Confirm name. Check collision. Create entity: name: "<Org Name> SWOT", entityType: "SWOT".

Mode Routing

Default add. Empty entity with review/challenge → redirect to add.

Add Mode (Capture)

Read capture-form.md for the form, parsing rules, and confirm flow.

Capture is observation-first. The user reports what they observed and against what baseline; the skill proposes the SWOT tag at the confirm step. The user never picks a SWOT bucket at input time — this removes the bucketing bias the challenge checks would otherwise have to clean up.

Conversational (default): 6 neutral prompts (what observed, how known, compared to what, what's missing, so-what, anything else) — none name a quadrant. Parse to draft observations, then propose [strength]/[weakness]/[opportunity]/[threat]/[context] plus a landscape tag at the confirm step, each with a one-line rationale. Confirm / Retag / Edit / Cancel → write one-at-a-time, fully tagged only (never persist an untagged observation, so the stored format matches prior entries). Failed writes → pending-sync. After write, offer challenge pass.

Artifact-pointed (--read): read file/URL, extract signals as draft observations, present in the same auto-tag confirm flow. Never writes unconfirmed observations.

Discovery handoff (--from=1on1-prep): see "Discovery Handoff" below.

Review Mode

Read entity. Render sections (omit empty): Header, Internal (strengths/weaknesses by landscape tag), External (opportunities/threats by landscape tag), Context, Coverage Gaps (4x4 SWOT × landscape matrix, flag cells <2 entries). Then offer export (see export-formats.md).

Challenge Mode

Read entity, run 4 checks (see challenge-checks.md). Present flagged items with Edit/Remove/Keep/Recategorize actions.

Discovery Handoff (--from=1on1-prep)

/1on1-prep is the discovery instrument; /swot is the sorter. This handoff carries captured 1:1 notes into a SWOT capture session so the strong, non-leading discovery questions reach the analysis without re-typing. Both skills write to the same knowledge graph, so this is a graph read — not a parser.

Scoping note: the graph has no Person→org relation (1on1-prep's only relation is reports_to, Person→Person — see ../1on1-prep/graph-schema.md). So the handoff cannot query "people at this org" directly. In an onboarding ramp the SWOT and the 1:1s belong to the same org, so the default is to pull from all Person entities; the user prunes at the confirm step. Accept optional person names to narrow it: --from=1on1-prep "Sarah Chen,Mike Lee".

1. Read 1:1 notes. mcp__memory__search_nodes to find Person entities (named ones, else all), then read their [1on1] observations — strategic-tagged entries especially ([opportunity], [concern], [relationship]), since those map most directly to SWOT signal.
2. Draft, don't write. Map each into a draft observation: the 1:1 text → observation body. 1on1-prep observations carry no provenance parenthetical, so synthesize (1:1 with <person>) from the source Person entity. Answers about the biggest risk, the most debt, or what's missing often seed the comparative / absence framing.
3. Confirm. Present the drafts in the same auto-tag confirm flow as conversational capture (proposed tag + one-line rationale). Nothing is written until the user confirms — same confirm-then-write invariant.

If there are no [1on1] observations yet, say so and fall back to conversational capture.

Integration Points (Stubbed)

--from=architecture-overview and --from=stakeholder-map are planned but unbuilt — those skills produce artifacts SWOT would have to interpret, unlike 1on1-prep which already writes tagged graph observations. If a user passes one of these --from values, return:

"The / skill isn't built yet. You can manually add insights using the conversational capture, or use --from=1on1-prep if you have 1:1 notes."

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):

• SWOT entities, relations, observations keyed on <Org Name> SWOT — cross-session structured graph state.

User working repo (ramp workspace):

• docs/swot/ markdown exports — committed deliverables.

Not used by this skill: auto-memory MD, ruflo MCP, scheduled-tasks MCP, plugin-internal memory (decisions.md / patterns.md).

Out-of-scope local files (not memory layers): pending-sync/ (transient fallback drained by --sync), Excalidraw / Slidev exports (rendering sinks, not persistence).

Common Mistakes

• Asking the user to pre-sort into SWOT buckets at capture time — capture is observation-first; gather what was observed and the baseline, then propose the tag at confirm. Naming quadrants in the capture prompts biases the answer.
• Persisting an untagged observation — tagging is deferred in the conversation only; every observation written to the graph carries its full [swot][landscape] shape, so review and challenge read new and old entries identically.
• Writing observations without provenance — every observation should carry its source (conversation, artifact, meeting); unattributed entries can't be audited or challenged later.
• Mis-tagging internal vs. external at confirm — strengths/weaknesses are internal to the org; opportunities/threats are external forces. The auto-tag proposal can get this wrong; the one-line rationale exists so the user catches it with Retag before writing.
• Writing observations without user confirmation — capture is confirm-then-write; never persist draft observations from artifact extraction or --from handoff until the user approves them.
• Reaching for challenge or review mode on an empty entity — redirect to add when the entity has no observations yet; there's nothing to review.
• Skipping coverage gaps in review — the 4x4 SWOT × landscape matrix is where missing analysis becomes visible; always surface cells with fewer than two entries.

Composition

• Reads: /1on1-prep (--from=1on1-prep, built), /architecture-overview, /stakeholder-map (future --from)
• Writes: knowledge graph, docs/swot/, excalidraw canvas, Slidev
• Feeds: /strategy-doc (#42), /okr (#36)
• Downstream review: this skill's Challenge Mode checks SWOT coverage; substance is stress-tested one level up — once these observations are synthesized into a strategy via /strategy-doc, that strategy gets the independent strategy-adversary pass. No adversary on raw observations.