Skills catalogue¶
The plugin's skills are grouped by the function that owns them — the six functions a think tank actually staffs. Inside each function, skills are ordered by stage: where they sit in the research workflow.
Grouping by function answers "which part of the organisation does this belong to", which is how you look for a skill you have not used before. The stage answers "when in the work do I reach for it", which is how you look for one mid-project.
Both lists are closed and registered in the plugin's controlled vocabulary, and every skill declares its own function and stage. This table is generated from those declarations, so it cannot drift from the catalogue that ships.
89 skills.
By intent — what brought you here?¶
- Need to frame a take, find a number, distil a source, or write and test a brief → Research
- Need to know what moved this week across institutions, media, industry or the literature → Policy intelligence
- Need to know who stands where on a file, or who to reach → Advocacy
- Need a deck, a paper, a carousel, a chart on the site → Communications
- Need a funder, a dossier on one, or the ask itself → Development
- Need to scaffold a project, audit a folder, build a skill, or just route → Operations
The workflow view¶
The same skills, seen by when in the work you reach for them. The phases aren't strictly sequential — you'll loop. Evidence often reshapes position, and quality gates apply at every step.
flowchart LR
SA["<b>Situation awareness</b><br/><i>transversal</i>"]
subgraph LOOP["sequential phases — iterate freely"]
direction LR
P["<b>Position</b><br/><b>formation</b>"]
E["<b>Evidence</b><br/><b>base</b>"]
O["<b>Output</b>"]
P --> E --> O
O -.->|iterate| P
E -.->|reshape| P
end
QG["<b>Quality gates</b><br/><i>cross-phase</i>"]
SUP["<b>Support</b><br/><i>transversal</i>"]
FUND["<b>Funding pipeline</b><br/><i>parallel track</i>"]
SA -->|feeds context| LOOP
QG -.->|reviews| LOOP
SUP -.->|scaffolds| LOOP
FUND -.->|funds| LOOPGaps we know about¶
Quality-gate skills we want and do not have: stakeholder-position-verifier, data-replication-check, tracker-design-review, embed-iframe-check, funder enum-saturation check.
On the roadmap: trilogue tracker · amendments diff · coalition and co-signatory mapping · options matrix · ECJ and case-law search · regulator decisions · impact-assessment review · scenario comparator · op-ed · talking points and meeting briefing · chart and visual review · translation EN↔FR/DE · grant reporting · glossary maintainer.
How you actually invoke one¶
You don't call skills by name. You describe what you want and Cowork routes you to the right skill based on trigger phrases declared in each SKILL.md. Examples:
- "draft a brief on capacity mechanisms" →
draft-brief - "what's new on the European Grids Package" →
search-eu-file - "find me the academic literature on carbon pricing pass-through" →
search-academic - "who are the rapporteurs on EPBD recast" →
search-eu-file
If you're not sure which skill applies, type help in any conversation — it routes you live.
Per-skill detail¶
Per-skill trigger phrases, full descriptions, and hand-offs live in each SKILL.md under plugins/empn-core/skills/. Or type help in Cowork — it enumerates the catalogue live and routes to the right one.
Trust and review¶
Skills are not magic. They encode methodology — what sources to consult, what folder to write to, what voice to use — and they keep the model honest by linking out to shared rules at runtime. They do not replace judgement. Treat any skill output as a draft you review; the quality-gate skills above (review-facts, review-adversarial) exists for exactly that purpose.
A scheduled CI agent runs consistency checks (audit-repo) and benchmark prompts (skill-benchmark) against every skill on a cadence. The aggregate status appears in the repo README.md; once the bench has run a few times, a live status table will replace the link on this page.