Agents
Route chart discussion, feedback, recommendations, and catalog improvement drafts through chartcoach skills.
chartcoach provides agent skills. The top-level $chartcoach skill routes
tasks to workflow skills bundled with the installed chartcoach package.
npx skills add chartcoach/skills
uvx chartcoach@latest skills listAsk for the job in the current agent session:
codex 'Use $chartcoach to review this dashboard screenshot for labeling, legend, and accessibility issues.'Skill paths
| Skill | Use it for |
|---|---|
core | Catalog source, output formats, command recovery, and index setup |
discuss | Cited design discussion, tradeoffs, comparisons, and source tracing |
visfeedback | Feedback on an observed chart or screenshot |
visrec | Visualization recommendations from a data task and constraints |
contribute | Local catalog improvement issue drafts |
Inspect bundled skill text:
uvx chartcoach@latest skills list --format jsonl
uvx chartcoach@latest skills get core
uvx chartcoach@latest skills get visfeedbackEvidence flow
Agents should keep retrieval and evidence separate:
catalog shape -> candidate ids -> exact guideline -> citation -> answered claimRead the live catalog shape with overview, manifest, roles, and labels.
Discover candidate ids with list, query, sql, or find. Verify each
candidate with catalog read, then produce links and source references with
catalog cite.
Search output is candidate evidence. Exact catalog read output is the text
that can support a claim.
uvx chartcoach@latest catalog query --label chart:bar:use --limit 5 --format jsonl
uvx chartcoach@latest catalog read compare-percentages-with-bars-not-pies \
--source-detail minimal \
--format json
uvx chartcoach@latest catalog cite compare-percentages-with-bars-not-pies \
--format jsonContribution drafts
Use contribute when retrieval exposes missing, weak, duplicated, or
hard-to-apply catalog guidance.
uvx chartcoach@latest skills get contributeThe skill drafts local Markdown for the chartcoach/catalog issue surface. It
does not post, submit, or disclose the draft unless a human approves that
external action.
Use MCP when an agent platform should call catalog SQL or search through tool invocations.