Present package install details via the view primitive instead of
embedding render.kv inside a prompt template string.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Add @release-it/bumper, @release-it/conventional-changelog, oxlint,
prettier, and release-it to the NOTICE file. All MIT-licensed.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
skill-kit 1.6.2 fixes nested branch reconvergence, so store.diagnosis
is now correctly inferred as guaranteed after the review step. Removes
all non-null assertions on the diagnosis sub-store.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
The store type system already infers step result types from action
output schemas and response schemas. Remove all redundant `as X`
casts on store.steps access. Only store.diagnosis! remains (the
reconvergence promotion doesn't yet track sub-store writes through
multi-hop branches — filed separately).
Also removes unused schema imports that were only needed for casts.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Align masking between env var table and credentials table — both now
show the start of the value (first 8 chars + ****). Space IDs and
environment names are shown in full since they aren't secrets.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
The confirm-credentials step now uses view() to render the detected
credentials table directly to the user, followed by act.askUser() for
the confirmation prompt. Previously the credentials were embedded in
the instruction prompt, so the agent would ask for confirmation without
first showing what was found.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Split the monolithic checkPackagesAndEnv action into two focused actions:
- checkPackages: scans package.json for installed SDKs and package manager
- scanCredentials: scans .env files for Contentful/Ninetailed tokens
Restructure the doctor subskill flow from a 4-way branch maze into
clean 2-way branches with reconvergence:
- New credential flow: explore → scan-credentials → confirm-credentials
→ apply-credentials (auto-advance) → check-api
- Bifurcation: credentials confirmed → full diagnostic with API checks
and optional entry inspection; no credentials → code-only analysis
- Both paths reconverge at review → diagnosis store guaranteed
- Merged report-only into done as universal terminal
Stores redesigned with separated concerns:
- project.packages (non-optional PackagesResult, not packageData blob)
- credentials with nested personalization/contentful objects (non-optional
on the credentials path via apply-credentials)
- diagnosis with non-optional fields (guaranteed via reconvergence)
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Adds .prettierrc (singleQuote, printWidth 120), .prettierignore, and
prettier as a devDependency. Reformats all source files to match.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Migrate from skill-kit v1.4 (stash-based state) to the new state system
redesign (store-based with DAG-aware type narrowing). This replaces the
unstructured stash with typed sub-stores that group data semantically and
decouple consumers from producer step names.
Key changes:
- Zod → ArkType for all schemas (schemas.ts + 6 action files)
- stash → sub-stores: onboard (project + setup), doctor (project +
credentials + diagnosis), develop (step access only)
- output → response, stepOutput → response, actionOutput → actionResult
- updateStash removed, replaced by save() callbacks that write to stores
- Subskill params mappers updated from (output, stash) → (output, store)
- Tests updated with valid ArkType-conformant mock data
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Plugin installs wire up MCP automatically, but users who installed via
npx skills add need to configure the two MCP servers themselves. Add
step-by-step instructions for both the Contentful MCP (HTTP) and the
personalization MCP (stdio) with Claude Code commands and generic JSON
config examples.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Lead with the Claude Code plugin install as the hero quickstart since it
provides the best experience (skills + MCP servers). Demote npx/universal
CLI to "Other platforms" section. Add scannable skills summary table,
stronger tagline, dark-mode-ready logo, and badges.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Plugin-bundled MCP tools use the format
mcp__plugin_contentful-skills_{server}__{tool} at runtime, which
doesn't match the standalone mcp__{server}__{tool} names. Add
wildcards for both formats so skills work without permission prompts
whether installed standalone or as a plugin.
Also grants all Contentful skills access to the Contentful MCP server.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
contentful-automation[bot]