chore: version packages

[photon-action-bot]

This PR was opened by the Changesets release GitHub action. When you're ready to do a release, you can merge this and the packages will be published to npm automatically. If you're not ready to do a release yet, that's fine, whenever you add more changesets to master, this PR will be updated.

Releases

@vellum-docs/cli@0.3.0

Minor Changes

• 2a9986c: Add vellum build --watch with per-template invalidation

  Edit a template or a source file and Vellum re-renders only the output files affected by the change. A docs project that wires vellum build --watch alongside the host's dev server (Next.js, Mintlify, etc.) no longer needs a restart loop to see TSDoc edits.

  vellum build --watch

  How invalidation works

  Templates are instrumented while rendering. The symbol(), symbols(), and module() globals record what each .vel file actually read:

  • symbol(id) - the looked-up SymbolId,
  • module(path) - the module path,
  • symbols(query) - the verbatim query and the resulting ids.

  On a source-file change, Vellum diffs the file's old vs new symbols (source-position-insensitive content hash) and re-renders templates whose recorded reads intersect the diff. Added symbols are re-matched against each template's prior queries so a new symbols({ tag: 'foo' }) result triggers the right renders.

  Template edits re-render only that template. Config changes tear down the watcher and re-prime. Render and extract errors log and keep the previous output on disk - the watcher stays alive.

  API changes (minor - pre-1.0)

  TemplateEngine.render returns RenderResult, not string. Custom engines update the return shape:

  // before
  async render(source, ctx): Promise<string> { ... }
  
  // after
  async render(source, ctx): Promise<RenderResult> {
    return { output }
    // or { output, reads } when ctx.reads was supplied
  }

  SymbolIndex gains symbolsByFile and removeByFile. Custom index implementations need both. InMemorySymbolIndex now tracks per-file provenance via Symbol.source.file.

  matchesQuery(sym, query) is exported from @vellum-docs/core - the exact predicate InMemorySymbolIndex.symbols() uses, reusable by tooling that needs to test a single symbol against a query.

  New exports (@vellum-docs/core)

  • TemplateReads, createTemplateReads()
  • SymbolDiff, diffSymbols, hashSymbol, mergeDiffs, emptyDiff, isEmptyDiff
  • DependencyGraph
  • Vellum.extractLanguage, Vellum.renderTemplate, Vellum.listTemplates

  New exports (@vellum-docs/cli)

  • runWatch, WatchCommandOptions

  Watcher details

  File watching uses chokidar v4 with Nuxt-style granular filtering: node_modules, .git, .turbo, dist, build, coverage, and the resolved outDir are ignored anywhere in the tree, even if a user's sources[lang].include would otherwise cover them. Events are debounced 100ms; awaitWriteFinish handles editor save races.

  Known tradeoff

  Per-file TS extraction isn't meaningful in isolation (cross-file type graph), so extractLanguage re-runs the extractor over the whole program on each batch. Invalidation still narrows the render set - which is what dominates latency for a typical docs project - but we don't yet skip extraction for unchanged files beyond the existing content-hash cache.

Patch Changes

• 3840674: Refactor CLI to citty + consola

  Replaced hand-rolled argv parsing and console.log with two unjs libraries:

  • citty drives argument parsing, subcommand routing, and help generation. vellum --help and vellum build --help now produce consistent typed output instead of a static string.
  • consola handles every user-facing log through a shared tagged logger. Output is color-coded by level (success, info, warn, error) with a uniform [vellum] prefix.

  No breaking CLI surface changes — vellum build, --watch, --config, --cwd, --no-strict all behave identically. --no-strict now comes from citty's boolean-negation convention instead of being parsed by hand.

• 6503a35: Philosophy audit fixes - strict-by-default, dead schema cleanup

  An audit against the newly-written PHILOSOPHY.md surfaced four gaps. This changeset closes them.

  Strict template rendering is now on by default. Principle 11 ("fail loudly at build time") was being violated by throwOnUndefined: false - a template with a typo ({{ fn.doc.summaryy }} instead of fn.doc.summary) silently rendered as empty string, and the docs shipped with a blank section. The NunjucksEngine now defaults to strict rendering: any output of an undefined value throws, which bubbles to a non-zero build exit.

  Opt-out paths, for the rare cases where silent fallback is wanted during migration:

  • Config: new NunjucksEngine({ strict: false }).
  • CLI: vellum build --no-strict.

  This is a behavior change. Templates that relied on silent-empty for undefined values will now fail. Typical patterns that are still safe: {% if sym.members %}, {{ sym.doc.summary }} (empty string is defined), {% for m in sym.members or [] %}. The patterns that will now break are the ones you wanted to know about anyway.

  Schema cleanup. Three dead schema fields removed - they were defined but never populated by any extractor, violating principle 7 ("80% case defines the schema"):

  • Symbol.signatureResolved?: string - removed.
  • Member.kind values 'index' and 'call' - removed from the union. Can be added back with implementation when a TS call/index-signature extractor lands or a language that needs them ships.

  Docs drift fixes. ARCHITECTURE.md referenced a {{ str | tsdoc }} filter that never existed; replaced with {{ sym | summary }} (which does). Principle 2 in PHILOSOPHY.md now explicitly distinguishes "pattern-aware" (OK) from "language-idiosyncratic" (not OK), so Symbol.discriminator? is consistent with the stated rule.

• Updated dependencies [2a9986c]

• Updated dependencies [ad1aa14]

• Updated dependencies [6503a35]

  • @vellum-docs/core@0.3.0

@vellum-docs/core@0.3.0

Minor Changes

• 2a9986c: Add vellum build --watch with per-template invalidation

  Edit a template or a source file and Vellum re-renders only the output files affected by the change. A docs project that wires vellum build --watch alongside the host's dev server (Next.js, Mintlify, etc.) no longer needs a restart loop to see TSDoc edits.

  vellum build --watch

  How invalidation works

  Templates are instrumented while rendering. The symbol(), symbols(), and module() globals record what each .vel file actually read:

  • symbol(id) - the looked-up SymbolId,
  • module(path) - the module path,
  • symbols(query) - the verbatim query and the resulting ids.

  On a source-file change, Vellum diffs the file's old vs new symbols (source-position-insensitive content hash) and re-renders templates whose recorded reads intersect the diff. Added symbols are re-matched against each template's prior queries so a new symbols({ tag: 'foo' }) result triggers the right renders.

  Template edits re-render only that template. Config changes tear down the watcher and re-prime. Render and extract errors log and keep the previous output on disk - the watcher stays alive.

  API changes (minor - pre-1.0)

  TemplateEngine.render returns RenderResult, not string. Custom engines update the return shape:

  // before
  async render(source, ctx): Promise<string> { ... }
  
  // after
  async render(source, ctx): Promise<RenderResult> {
    return { output }
    // or { output, reads } when ctx.reads was supplied
  }

  SymbolIndex gains symbolsByFile and removeByFile. Custom index implementations need both. InMemorySymbolIndex now tracks per-file provenance via Symbol.source.file.

  matchesQuery(sym, query) is exported from @vellum-docs/core - the exact predicate InMemorySymbolIndex.symbols() uses, reusable by tooling that needs to test a single symbol against a query.

  New exports (@vellum-docs/core)

  • TemplateReads, createTemplateReads()
  • SymbolDiff, diffSymbols, hashSymbol, mergeDiffs, emptyDiff, isEmptyDiff
  • DependencyGraph
  • Vellum.extractLanguage, Vellum.renderTemplate, Vellum.listTemplates

  New exports (@vellum-docs/cli)

  • runWatch, WatchCommandOptions

  Watcher details

  File watching uses chokidar v4 with Nuxt-style granular filtering: node_modules, .git, .turbo, dist, build, coverage, and the resolved outDir are ignored anywhere in the tree, even if a user's sources[lang].include would otherwise cover them. Events are debounced 100ms; awaitWriteFinish handles editor save races.

  Known tradeoff

  Per-file TS extraction isn't meaningful in isolation (cross-file type graph), so extractLanguage re-runs the extractor over the whole program on each batch. Invalidation still narrows the render set - which is what dominates latency for a typical docs project - but we don't yet skip extraction for unchanged files beyond the existing content-hash cache.

Patch Changes

• ad1aa14: Add cell filter + TypeString.oneline for cell-safe rendering

  Every adopter was hand-rolling the same 5-filter escape chain when dropping types or summaries into markdown table cells:

  {{ m.type.text | replace("\n"," ") | replace("    ","") | replace("|","\\|") | replace("<","&lt;") | replace(">","&gt;") }}

  Two additions collapse that:

  TypeString.oneline?: string - populated at extraction time with the whitespace-collapsed form of text. Omitted when equal to text (single-line case). Fixes the \n + indentation problem at source, before any template filter runs.

  cell filter (profile-routed) - accepts a TypeString, plain string, or null. Collapses whitespace as defence-in-depth, routes through the profile's new cell(value, ctx) method, which wraps in a code span and escapes |. Works for anything cell-bound, not just types.

  Before:

  | `{{ m.name }}` | `{{ m.type.text | replace("\n"," ") | replace("    ","") | replace("|","\\|") | replace("<","&lt;") | replace(">","&gt;") }}` | {{ m.doc.summary }} |

  After:

  | `{{ m.name }}` | {{ m.type | cell | safe }} | {{ m.doc.summary | cell | safe }} |

  Schema additions (additive)

  • TypeString.oneline?: string
  • RendererProfile.cell(value: string, ctx: RenderContext): string

  Existing extractors keep working - when oneline is absent the filter falls back to .text. Existing profiles get the new method implemented in MarkdownProfile and MintlifyProfile; third-party profiles must add a cell implementation.

  Out of scope

  jsx-prop and fenced contexts from the original request. Neither has recurring template pain today; defer until they do.

• 6503a35: Philosophy audit fixes - strict-by-default, dead schema cleanup

  An audit against the newly-written PHILOSOPHY.md surfaced four gaps. This changeset closes them.

  Strict template rendering is now on by default. Principle 11 ("fail loudly at build time") was being violated by throwOnUndefined: false - a template with a typo ({{ fn.doc.summaryy }} instead of fn.doc.summary) silently rendered as empty string, and the docs shipped with a blank section. The NunjucksEngine now defaults to strict rendering: any output of an undefined value throws, which bubbles to a non-zero build exit.

  Opt-out paths, for the rare cases where silent fallback is wanted during migration:

  • Config: new NunjucksEngine({ strict: false }).
  • CLI: vellum build --no-strict.

  This is a behavior change. Templates that relied on silent-empty for undefined values will now fail. Typical patterns that are still safe: {% if sym.members %}, {{ sym.doc.summary }} (empty string is defined), {% for m in sym.members or [] %}. The patterns that will now break are the ones you wanted to know about anyway.

  Schema cleanup. Three dead schema fields removed - they were defined but never populated by any extractor, violating principle 7 ("80% case defines the schema"):

  • Symbol.signatureResolved?: string - removed.
  • Member.kind values 'index' and 'call' - removed from the union. Can be added back with implementation when a TS call/index-signature extractor lands or a language that needs them ships.

  Docs drift fixes. ARCHITECTURE.md referenced a {{ str | tsdoc }} filter that never existed; replaced with {{ sym | summary }} (which does). Principle 2 in PHILOSOPHY.md now explicitly distinguishes "pattern-aware" (OK) from "language-idiosyncratic" (not OK), so Symbol.discriminator? is consistent with the stated rule.

@vellum-docs/engine-nunjucks@0.3.0

Minor Changes

• 2a9986c: Add vellum build --watch with per-template invalidation

  Edit a template or a source file and Vellum re-renders only the output files affected by the change. A docs project that wires vellum build --watch alongside the host's dev server (Next.js, Mintlify, etc.) no longer needs a restart loop to see TSDoc edits.

  vellum build --watch

  How invalidation works

  Templates are instrumented while rendering. The symbol(), symbols(), and module() globals record what each .vel file actually read:

  • symbol(id) - the looked-up SymbolId,
  • module(path) - the module path,
  • symbols(query) - the verbatim query and the resulting ids.

  On a source-file change, Vellum diffs the file's old vs new symbols (source-position-insensitive content hash) and re-renders templates whose recorded reads intersect the diff. Added symbols are re-matched against each template's prior queries so a new symbols({ tag: 'foo' }) result triggers the right renders.

  Template edits re-render only that template. Config changes tear down the watcher and re-prime. Render and extract errors log and keep the previous output on disk - the watcher stays alive.

  API changes (minor - pre-1.0)

  TemplateEngine.render returns RenderResult, not string. Custom engines update the return shape:

  // before
  async render(source, ctx): Promise<string> { ... }
  
  // after
  async render(source, ctx): Promise<RenderResult> {
    return { output }
    // or { output, reads } when ctx.reads was supplied
  }

  SymbolIndex gains symbolsByFile and removeByFile. Custom index implementations need both. InMemorySymbolIndex now tracks per-file provenance via Symbol.source.file.

  matchesQuery(sym, query) is exported from @vellum-docs/core - the exact predicate InMemorySymbolIndex.symbols() uses, reusable by tooling that needs to test a single symbol against a query.

  New exports (@vellum-docs/core)

  • TemplateReads, createTemplateReads()
  • SymbolDiff, diffSymbols, hashSymbol, mergeDiffs, emptyDiff, isEmptyDiff
  • DependencyGraph
  • Vellum.extractLanguage, Vellum.renderTemplate, Vellum.listTemplates

  New exports (@vellum-docs/cli)

  • runWatch, WatchCommandOptions

  Watcher details

  File watching uses chokidar v4 with Nuxt-style granular filtering: node_modules, .git, .turbo, dist, build, coverage, and the resolved outDir are ignored anywhere in the tree, even if a user's sources[lang].include would otherwise cover them. Events are debounced 100ms; awaitWriteFinish handles editor save races.

  Known tradeoff

  Per-file TS extraction isn't meaningful in isolation (cross-file type graph), so extractLanguage re-runs the extractor over the whole program on each batch. Invalidation still narrows the render set - which is what dominates latency for a typical docs project - but we don't yet skip extraction for unchanged files beyond the existing content-hash cache.

Patch Changes

• ad1aa14: Add cell filter + TypeString.oneline for cell-safe rendering

  Every adopter was hand-rolling the same 5-filter escape chain when dropping types or summaries into markdown table cells:

  {{ m.type.text | replace("\n"," ") | replace("    ","") | replace("|","\\|") | replace("<","&lt;") | replace(">","&gt;") }}

  Two additions collapse that:

  TypeString.oneline?: string - populated at extraction time with the whitespace-collapsed form of text. Omitted when equal to text (single-line case). Fixes the \n + indentation problem at source, before any template filter runs.

  cell filter (profile-routed) - accepts a TypeString, plain string, or null. Collapses whitespace as defence-in-depth, routes through the profile's new cell(value, ctx) method, which wraps in a code span and escapes |. Works for anything cell-bound, not just types.

  Before:

  | `{{ m.name }}` | `{{ m.type.text | replace("\n"," ") | replace("    ","") | replace("|","\\|") | replace("<","&lt;") | replace(">","&gt;") }}` | {{ m.doc.summary }} |

  After:

  | `{{ m.name }}` | {{ m.type | cell | safe }} | {{ m.doc.summary | cell | safe }} |

  Schema additions (additive)

  • TypeString.oneline?: string
  • RendererProfile.cell(value: string, ctx: RenderContext): string

  Existing extractors keep working - when oneline is absent the filter falls back to .text. Existing profiles get the new method implemented in MarkdownProfile and MintlifyProfile; third-party profiles must add a cell implementation.

  Out of scope

  jsx-prop and fenced contexts from the original request. Neither has recurring template pain today; defer until they do.

• 6503a35: Philosophy audit fixes - strict-by-default, dead schema cleanup

  An audit against the newly-written PHILOSOPHY.md surfaced four gaps. This changeset closes them.

  Strict template rendering is now on by default. Principle 11 ("fail loudly at build time") was being violated by throwOnUndefined: false - a template with a typo ({{ fn.doc.summaryy }} instead of fn.doc.summary) silently rendered as empty string, and the docs shipped with a blank section. The NunjucksEngine now defaults to strict rendering: any output of an undefined value throws, which bubbles to a non-zero build exit.

  Opt-out paths, for the rare cases where silent fallback is wanted during migration:

  • Config: new NunjucksEngine({ strict: false }).
  • CLI: vellum build --no-strict.

  This is a behavior change. Templates that relied on silent-empty for undefined values will now fail. Typical patterns that are still safe: {% if sym.members %}, {{ sym.doc.summary }} (empty string is defined), {% for m in sym.members or [] %}. The patterns that will now break are the ones you wanted to know about anyway.

  Schema cleanup. Three dead schema fields removed - they were defined but never populated by any extractor, violating principle 7 ("80% case defines the schema"):

  • Symbol.signatureResolved?: string - removed.
  • Member.kind values 'index' and 'call' - removed from the union. Can be added back with implementation when a TS call/index-signature extractor lands or a language that needs them ships.

  Docs drift fixes. ARCHITECTURE.md referenced a {{ str | tsdoc }} filter that never existed; replaced with {{ sym | summary }} (which does). Principle 2 in PHILOSOPHY.md now explicitly distinguishes "pattern-aware" (OK) from "language-idiosyncratic" (not OK), so Symbol.discriminator? is consistent with the stated rule.

• Updated dependencies [2a9986c]

• Updated dependencies [ad1aa14]

• Updated dependencies [6503a35]

  • @vellum-docs/core@0.3.0

@vellum-docs/extractor-typescript@0.3.0

Patch Changes

• ad1aa14: Add cell filter + TypeString.oneline for cell-safe rendering

  Every adopter was hand-rolling the same 5-filter escape chain when dropping types or summaries into markdown table cells:

  {{ m.type.text | replace("\n"," ") | replace("    ","") | replace("|","\\|") | replace("<","&lt;") | replace(">","&gt;") }}

  Two additions collapse that:

  TypeString.oneline?: string - populated at extraction time with the whitespace-collapsed form of text. Omitted when equal to text (single-line case). Fixes the \n + indentation problem at source, before any template filter runs.

  cell filter (profile-routed) - accepts a TypeString, plain string, or null. Collapses whitespace as defence-in-depth, routes through the profile's new cell(value, ctx) method, which wraps in a code span and escapes |. Works for anything cell-bound, not just types.

  Before:

  | `{{ m.name }}` | `{{ m.type.text | replace("\n"," ") | replace("    ","") | replace("|","\\|") | replace("<","&lt;") | replace(">","&gt;") }}` | {{ m.doc.summary }} |

  After:

  | `{{ m.name }}` | {{ m.type | cell | safe }} | {{ m.doc.summary | cell | safe }} |

  Schema additions (additive)

  • TypeString.oneline?: string
  • RendererProfile.cell(value: string, ctx: RenderContext): string

  Existing extractors keep working - when oneline is absent the filter falls back to .text. Existing profiles get the new method implemented in MarkdownProfile and MintlifyProfile; third-party profiles must add a cell implementation.

  Out of scope

  jsx-prop and fenced contexts from the original request. Neither has recurring template pain today; defer until they do.

• 6503a35: Philosophy audit fixes - strict-by-default, dead schema cleanup

  An audit against the newly-written PHILOSOPHY.md surfaced four gaps. This changeset closes them.

  Strict template rendering is now on by default. Principle 11 ("fail loudly at build time") was being violated by throwOnUndefined: false - a template with a typo ({{ fn.doc.summaryy }} instead of fn.doc.summary) silently rendered as empty string, and the docs shipped with a blank section. The NunjucksEngine now defaults to strict rendering: any output of an undefined value throws, which bubbles to a non-zero build exit.

  Opt-out paths, for the rare cases where silent fallback is wanted during migration:

  • Config: new NunjucksEngine({ strict: false }).
  • CLI: vellum build --no-strict.

  This is a behavior change. Templates that relied on silent-empty for undefined values will now fail. Typical patterns that are still safe: {% if sym.members %}, {{ sym.doc.summary }} (empty string is defined), {% for m in sym.members or [] %}. The patterns that will now break are the ones you wanted to know about anyway.

  Schema cleanup. Three dead schema fields removed - they were defined but never populated by any extractor, violating principle 7 ("80% case defines the schema"):

  • Symbol.signatureResolved?: string - removed.
  • Member.kind values 'index' and 'call' - removed from the union. Can be added back with implementation when a TS call/index-signature extractor lands or a language that needs them ships.

  Docs drift fixes. ARCHITECTURE.md referenced a {{ str | tsdoc }} filter that never existed; replaced with {{ sym | summary }} (which does). Principle 2 in PHILOSOPHY.md now explicitly distinguishes "pattern-aware" (OK) from "language-idiosyncratic" (not OK), so Symbol.discriminator? is consistent with the stated rule.

• Updated dependencies [2a9986c]

• Updated dependencies [ad1aa14]

• Updated dependencies [6503a35]

  • @vellum-docs/core@0.3.0

@vellum-docs/language-server@0.3.0

Patch Changes

• ad1aa14: Add cell filter + TypeString.oneline for cell-safe rendering

  Every adopter was hand-rolling the same 5-filter escape chain when dropping types or summaries into markdown table cells:

  {{ m.type.text | replace("\n"," ") | replace("    ","") | replace("|","\\|") | replace("<","&lt;") | replace(">","&gt;") }}

  Two additions collapse that:

  TypeString.oneline?: string - populated at extraction time with the whitespace-collapsed form of text. Omitted when equal to text (single-line case). Fixes the \n + indentation problem at source, before any template filter runs.

  cell filter (profile-routed) - accepts a TypeString, plain string, or null. Collapses whitespace as defence-in-depth, routes through the profile's new cell(value, ctx) method, which wraps in a code span and escapes |. Works for anything cell-bound, not just types.

  Before:

  | `{{ m.name }}` | `{{ m.type.text | replace("\n"," ") | replace("    ","") | replace("|","\\|") | replace("<","&lt;") | replace(">","&gt;") }}` | {{ m.doc.summary }} |

  After:

  | `{{ m.name }}` | {{ m.type | cell | safe }} | {{ m.doc.summary | cell | safe }} |

  Schema additions (additive)

  • TypeString.oneline?: string
  • RendererProfile.cell(value: string, ctx: RenderContext): string

  Existing extractors keep working - when oneline is absent the filter falls back to .text. Existing profiles get the new method implemented in MarkdownProfile and MintlifyProfile; third-party profiles must add a cell implementation.

  Out of scope

  jsx-prop and fenced contexts from the original request. Neither has recurring template pain today; defer until they do.

• 6503a35: Philosophy audit fixes - strict-by-default, dead schema cleanup

  An audit against the newly-written PHILOSOPHY.md surfaced four gaps. This changeset closes them.

  Strict template rendering is now on by default. Principle 11 ("fail loudly at build time") was being violated by throwOnUndefined: false - a template with a typo ({{ fn.doc.summaryy }} instead of fn.doc.summary) silently rendered as empty string, and the docs shipped with a blank section. The NunjucksEngine now defaults to strict rendering: any output of an undefined value throws, which bubbles to a non-zero build exit.

  Opt-out paths, for the rare cases where silent fallback is wanted during migration:

  • Config: new NunjucksEngine({ strict: false }).
  • CLI: vellum build --no-strict.

  This is a behavior change. Templates that relied on silent-empty for undefined values will now fail. Typical patterns that are still safe: {% if sym.members %}, {{ sym.doc.summary }} (empty string is defined), {% for m in sym.members or [] %}. The patterns that will now break are the ones you wanted to know about anyway.

  Schema cleanup. Three dead schema fields removed - they were defined but never populated by any extractor, violating principle 7 ("80% case defines the schema"):

  • Symbol.signatureResolved?: string - removed.
  • Member.kind values 'index' and 'call' - removed from the union. Can be added back with implementation when a TS call/index-signature extractor lands or a language that needs them ships.

  Docs drift fixes. ARCHITECTURE.md referenced a {{ str | tsdoc }} filter that never existed; replaced with {{ sym | summary }} (which does). Principle 2 in PHILOSOPHY.md now explicitly distinguishes "pattern-aware" (OK) from "language-idiosyncratic" (not OK), so Symbol.discriminator? is consistent with the stated rule.

• Updated dependencies [2a9986c]

• Updated dependencies [ad1aa14]

• Updated dependencies [6503a35]

  • @vellum-docs/core@0.3.0

@vellum-docs/profile-markdown@0.3.0

Patch Changes

• ad1aa14: Add cell filter + TypeString.oneline for cell-safe rendering

  Every adopter was hand-rolling the same 5-filter escape chain when dropping types or summaries into markdown table cells:

  {{ m.type.text | replace("\n"," ") | replace("    ","") | replace("|","\\|") | replace("<","&lt;") | replace(">","&gt;") }}

  Two additions collapse that:

  TypeString.oneline?: string - populated at extraction time with the whitespace-collapsed form of text. Omitted when equal to text (single-line case). Fixes the \n + indentation problem at source, before any template filter runs.

  cell filter (profile-routed) - accepts a TypeString, plain string, or null. Collapses whitespace as defence-in-depth, routes through the profile's new cell(value, ctx) method, which wraps in a code span and escapes |. Works for anything cell-bound, not just types.

  Before:

  | `{{ m.name }}` | `{{ m.type.text | replace("\n"," ") | replace("    ","") | replace("|","\\|") | replace("<","&lt;") | replace(">","&gt;") }}` | {{ m.doc.summary }} |

  After:

  | `{{ m.name }}` | {{ m.type | cell | safe }} | {{ m.doc.summary | cell | safe }} |

  Schema additions (additive)

  • TypeString.oneline?: string
  • RendererProfile.cell(value: string, ctx: RenderContext): string

  Existing extractors keep working - when oneline is absent the filter falls back to .text. Existing profiles get the new method implemented in MarkdownProfile and MintlifyProfile; third-party profiles must add a cell implementation.

  Out of scope

  jsx-prop and fenced contexts from the original request. Neither has recurring template pain today; defer until they do.

• 6503a35: Philosophy audit fixes - strict-by-default, dead schema cleanup

  An audit against the newly-written PHILOSOPHY.md surfaced four gaps. This changeset closes them.

  Strict template rendering is now on by default. Principle 11 ("fail loudly at build time") was being violated by throwOnUndefined: false - a template with a typo ({{ fn.doc.summaryy }} instead of fn.doc.summary) silently rendered as empty string, and the docs shipped with a blank section. The NunjucksEngine now defaults to strict rendering: any output of an undefined value throws, which bubbles to a non-zero build exit.

  Opt-out paths, for the rare cases where silent fallback is wanted during migration:

  • Config: new NunjucksEngine({ strict: false }).
  • CLI: vellum build --no-strict.

  This is a behavior change. Templates that relied on silent-empty for undefined values will now fail. Typical patterns that are still safe: {% if sym.members %}, {{ sym.doc.summary }} (empty string is defined), {% for m in sym.members or [] %}. The patterns that will now break are the ones you wanted to know about anyway.

  Schema cleanup. Three dead schema fields removed - they were defined but never populated by any extractor, violating principle 7 ("80% case defines the schema"):

  • Symbol.signatureResolved?: string - removed.
  • Member.kind values 'index' and 'call' - removed from the union. Can be added back with implementation when a TS call/index-signature extractor lands or a language that needs them ships.

  Docs drift fixes. ARCHITECTURE.md referenced a {{ str | tsdoc }} filter that never existed; replaced with {{ sym | summary }} (which does). Principle 2 in PHILOSOPHY.md now explicitly distinguishes "pattern-aware" (OK) from "language-idiosyncratic" (not OK), so Symbol.discriminator? is consistent with the stated rule.

• Updated dependencies [2a9986c]

• Updated dependencies [ad1aa14]

• Updated dependencies [6503a35]

  • @vellum-docs/core@0.3.0

@vellum-docs/profile-mintlify@0.3.0

Patch Changes

• ad1aa14: Add cell filter + TypeString.oneline for cell-safe rendering

  Every adopter was hand-rolling the same 5-filter escape chain when dropping types or summaries into markdown table cells:

  {{ m.type.text | replace("\n"," ") | replace("    ","") | replace("|","\\|") | replace("<","&lt;") | replace(">","&gt;") }}

  Two additions collapse that:

  TypeString.oneline?: string - populated at extraction time with the whitespace-collapsed form of text. Omitted when equal to text (single-line case). Fixes the \n + indentation problem at source, before any template filter runs.

  cell filter (profile-routed) - accepts a TypeString, plain string, or null. Collapses whitespace as defence-in-depth, routes through the profile's new cell(value, ctx) method, which wraps in a code span and escapes |. Works for anything cell-bound, not just types.

  Before:

  | `{{ m.name }}` | `{{ m.type.text | replace("\n"," ") | replace("    ","") | replace("|","\\|") | replace("<","&lt;") | replace(">","&gt;") }}` | {{ m.doc.summary }} |

  After:

  | `{{ m.name }}` | {{ m.type | cell | safe }} | {{ m.doc.summary | cell | safe }} |

  Schema additions (additive)

  • TypeString.oneline?: string
  • RendererProfile.cell(value: string, ctx: RenderContext): string

  Existing extractors keep working - when oneline is absent the filter falls back to .text. Existing profiles get the new method implemented in MarkdownProfile and MintlifyProfile; third-party profiles must add a cell implementation.

  Out of scope

  jsx-prop and fenced contexts from the original request. Neither has recurring template pain today; defer until they do.

• 6503a35: Philosophy audit fixes - strict-by-default, dead schema cleanup

  An audit against the newly-written PHILOSOPHY.md surfaced four gaps. This changeset closes them.

  Strict template rendering is now on by default. Principle 11 ("fail loudly at build time") was being violated by throwOnUndefined: false - a template with a typo ({{ fn.doc.summaryy }} instead of fn.doc.summary) silently rendered as empty string, and the docs shipped with a blank section. The NunjucksEngine now defaults to strict rendering: any output of an undefined value throws, which bubbles to a non-zero build exit.

  Opt-out paths, for the rare cases where silent fallback is wanted during migration:

  • Config: new NunjucksEngine({ strict: false }).
  • CLI: vellum build --no-strict.

  This is a behavior change. Templates that relied on silent-empty for undefined values will now fail. Typical patterns that are still safe: {% if sym.members %}, {{ sym.doc.summary }} (empty string is defined), {% for m in sym.members or [] %}. The patterns that will now break are the ones you wanted to know about anyway.

  Schema cleanup. Three dead schema fields removed - they were defined but never populated by any extractor, violating principle 7 ("80% case defines the schema"):

  • Symbol.signatureResolved?: string - removed.
  • Member.kind values 'index' and 'call' - removed from the union. Can be added back with implementation when a TS call/index-signature extractor lands or a language that needs them ships.

  Docs drift fixes. ARCHITECTURE.md referenced a {{ str | tsdoc }} filter that never existed; replaced with {{ sym | summary }} (which does). Principle 2 in PHILOSOPHY.md now explicitly distinguishes "pattern-aware" (OK) from "language-idiosyncratic" (not OK), so Symbol.discriminator? is consistent with the stated rule.

• Updated dependencies [2a9986c]

• Updated dependencies [ad1aa14]

• Updated dependencies [6503a35]

  • @vellum-docs/core@0.3.0

vellum-example-basic@0.0.8

Patch Changes

• Updated dependencies [2a9986c]
• Updated dependencies [ad1aa14]
• Updated dependencies [3840674]
• Updated dependencies [6503a35]
  • @vellum-docs/cli@0.3.0
  • @vellum-docs/core@0.3.0
  • @vellum-docs/engine-nunjucks@0.3.0
  • @vellum-docs/extractor-typescript@0.3.0
  • @vellum-docs/profile-markdown@0.3.0
  • @vellum-docs/profile-mintlify@0.3.0

vellum-example-nextjs@0.0.8

Patch Changes

• Updated dependencies [2a9986c]
• Updated dependencies [ad1aa14]
• Updated dependencies [3840674]
• Updated dependencies [6503a35]
  • @vellum-docs/cli@0.3.0
  • @vellum-docs/core@0.3.0
  • @vellum-docs/engine-nunjucks@0.3.0
  • @vellum-docs/extractor-typescript@0.3.0
  • @vellum-docs/profile-markdown@0.3.0

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
vellum-docs	Ready	Preview, Comment	Apr 15, 2026 3:32am