Envisioning skills

Foresight methods, packaged as skills. Each one encodes a procedure we run on client work, in the order the loop actually runs:

stage	question	skills
scope	what decision, what boundary?	scope
sense	what is changing?	signal · ground · the seven scans below
structure	what does it connect to?	merge · frames · implications
project	where could it go?	scenarios
stress	what does my plan assume?	grill-the-future
make	what do I hand someone?	report · artifact-from-the-future

The seven scans

Each runs the same pipeline — generate → merge → ground → score — against a playbook that fixes the boundary, categories, metrics and sourcing rules. These are the playbooks we run for clients in Signals AI.

skill	boundary you give it	categories	scored on
horizon-scan	a topic or domain	STEEP	impact
technology-scan	a field, sector or region	hardware · software · applications · ethics & security	readiness · impact · investment
innovation-radar	an industry or function	business model · operating model · process · product · channel · CX	novelty · adoption momentum · fit
competitive-scan	a named competitor set	product · pricing · GTM · hiring · funding · messaging	threat · momentum · confidence
adjacency-scan	your home position	category · customer · geography · channel · capability	fit · defensibility · time to entry
regulatory-scan	a jurisdiction + topic	proposed · enacted · enforcement · self-reg · court · standards	compliance impact · time to effect · breadth
risk-radar	an exposure surface	operational · geopolitical · supply chain · cyber · climate · reputational	likelihood · severity · horizon

They chain. scope writes the briefing; scans emit cards; ground makes them traceable; merge and frames structure them; scenarios, implications and grill-the-future use them as evidence; report and artifact-from-the-future turn them into something you can hand over.

Every card carries a grounding state and per-source verdicts (core/grounding.md) — a signal is never asserted without saying what was read and what it said.

See it first

examples/europes-space-economy/ is a real end-to-end run — briefing, scan index, three grounded signal cards, four tensions, and the readout — with the coverage gaps it actually had.

Install

claude plugin marketplace add envisioning/skills
claude plugin install envisioning

or, without Claude Code: copy skills/<name>/SKILL.md and core/ into your agent's skills directory. Every skill reads core/ by relative path.

Standalone vs. connected

Every skill works with no account and no network. Outputs land in ./foresight/ in your project, as markdown with frontmatter.

When an Envisioning MCP is connected (search, open, docs_write are present), the same skills:

• ground — check what Envisioning already knows before generating (search for prior signals, methodology, taxonomy entries)
• persist — write the output into Docs at a predictable path with the right category, so it is findable, annotatable and shareable
• classify — map to the Envisioning technology taxonomy instead of the generic STEEP fallback

The contract is in core/envisioning.md. The same markdown file is what gets written either way; the MCP only changes where.

Layout

core/          shared references every skill reads — never invoked directly
  signal-schema.md   the one card format
  taxonomy.md        STEEP fallback + how to resolve the live taxonomy
  maturity.md        readiness scale
  voice.md           house style
  envisioning.md     ground / persist protocol
  grounding.md       grounding states, source verdicts, override rules
  scan.md            the generate → merge → ground → score pipeline
  playbooks/<slug>.md  one per scan: boundary, categories, metrics, sourcing
skills/<name>/SKILL.md

Skills are flat on disk because that is what plugin loaders discover; the stage lives in each skill's frontmatter and in this table.