fix: use project-root-relative paths in skill files (#2394)

Skills run from the project root, not their own directory. Bare
references/ and scripts/ paths wouldn't resolve. Use .agents/skills/
prefixed paths so scripts and reference docs are found correctly.

Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: TC-MO

.agents/skills/api-doc/SKILL.md

@@ -23,5 +23,5 @@ argument-hint: endpoint-name
 
 **CRITICAL**: API docs are generated, not hand-written. Never edit files in `apify-api/docs/` directly.
 
-For detailed patterns and examples, see `references/openapi-patterns.md`.
-For edge cases and process notes, see `references/process.md`.
+For detailed patterns and examples, see `.agents/skills/api-doc/references/openapi-patterns.md`.
+For edge cases and process notes, see `.agents/skills/api-doc/references/process.md`.

.agents/skills/api-doc/references/process.md

@@ -26,7 +26,7 @@ apify-api/openapi/
 
 ## Step 3: Create or update files
 
-See `references/openapi-patterns.md` for detailed patterns and examples.
+See `.agents/skills/api-doc/references/openapi-patterns.md` for detailed patterns and examples.
 
 Key conventions:
 

.agents/skills/doc-write/SKILL.md

@@ -20,5 +20,5 @@ argument-hint: topic
 
 Key rules: US English, active voice, imperative tone, sentence case headings, bold for UI elements only, all admonitions need titles, code blocks need language tags.
 
-For detailed structure templates, see `references/doc-structures.md`.
-For edge cases and process notes, see `references/process.md`.
+For detailed structure templates, see `.agents/skills/doc-write/references/doc-structures.md`.
+For edge cases and process notes, see `.agents/skills/doc-write/references/process.md`.

.agents/skills/doc-write/references/process.md

@@ -30,7 +30,7 @@ slug: /path/to/page
 
 ## Step 4: Write content
 
-Follow the structure template for the doc type in `references/doc-structures.md`:
+Follow the structure template for the doc type in `.agents/skills/doc-write/references/doc-structures.md`:
 
 - **Platform docs**: introduction → when to use → configure/use → best practices → related features
 - **Guides**: introduction → prerequisites → step-by-step → code examples → testing → summary

.agents/skills/review-docs/SKILL.md

@@ -14,7 +14,7 @@ argument-hint: file-path
 2. **Run deterministic checks** (main process) - these are objective, no judgment needed:
    - `npm run lint:md` (heading hierarchy, list numbering, spacing)
    - `vale "<file>" --minAlertLevel=error` (prose style, pronouns, dashes, code fences, admonitions)
-   - `scripts/check-frontmatter.sh "<file>"` (description char count)
+   - `.agents/skills/review-docs/scripts/check-frontmatter.sh "<file>"` (description char count)
 3. **Delegated standards review** - spawn one subagent per standards file to check compliance. Each subagent reads the file being reviewed plus one standards file, and returns violations with line numbers and suggested fixes:
    - Subagent 1: check against `standards/writing-style.md` (voice, tone, headings, links)
    - Subagent 2: check against `standards/content-standards.md` (front matter, admonitions, code blocks)
@@ -26,8 +26,8 @@ argument-hint: file-path
    - Technical accuracy (code examples correct, API endpoints current)
    - Completeness (prerequisites listed, edge cases covered)
    - Code example quality (complete, runnable, commented where needed)
-5. **Format output** - merge subagent findings + deterministic results + content review per `references/review-format.md`
+5. **Format output** - merge subagent findings + deterministic results + content review per `.agents/skills/review-docs/references/review-format.md`
 
 Deterministic tools first, then delegated standards checks, then content review. Report tool failures as objective facts. Report standards and content findings as judgment calls.
 
-For detailed process notes and edge cases, see `references/process.md`.
+For detailed process notes and edge cases, see `.agents/skills/review-docs/references/process.md`.

.agents/skills/review-docs/references/process.md

@@ -13,7 +13,7 @@ These are objective - no judgment needed. Report all failures. Run in the main p
 
 - `npm run lint:md` (markdownlint: heading hierarchy, double spaces, list numbering)
 - `vale "<file>" --minAlertLevel=error` (prose style, dashes, code fences, admonitions)
-- `scripts/check-frontmatter.sh "<file>"` (description char count)
+- `.agents/skills/review-docs/scripts/check-frontmatter.sh "<file>"` (description char count)
 
 ## Step 3: Delegated standards review
 
@@ -41,7 +41,7 @@ Run in the main process. Focus on what neither deterministic tools nor standards
 
 ## Step 5: Format output
 
-Merge deterministic results + subagent findings + content review into structured output per `references/review-format.md`.
+Merge deterministic results + subagent findings + content review into structured output per `.agents/skills/review-docs/references/review-format.md`.
 
 - Tool findings are objective facts
 - Standards findings are rule-based judgment calls

.agents/skills/tutorial/SKILL.md

@@ -25,5 +25,5 @@ argument-hint: topic
 
 Each step should have a clear action verb, expected result, and verification. Code examples must be complete - no pseudocode.
 
-For the detailed 8-section template, see `references/tutorial-template.md`.
-For edge cases and process notes, see `references/process.md`.
+For the detailed 8-section template, see `.agents/skills/tutorial/references/tutorial-template.md`.
+For edge cases and process notes, see `.agents/skills/tutorial/references/process.md`.

.agents/skills/tutorial/references/process.md

@@ -29,7 +29,7 @@ Use simple present tense ("Create an Actor" not "Creating an Actor"). Match slug
 
 ## Step 4: Write tutorial content
 
-Follow the 8-section template in `references/tutorial-template.md`:
+Follow the 8-section template in `.agents/skills/tutorial/references/tutorial-template.md`:
 
 1. Introduction with learning objectives
 1. Prerequisites with checklist and time estimate
@@ -56,7 +56,7 @@ Verify front matter has all required fields. Run `npm run build` to catch broken
 
 ### Code examples don't render correctly
 
-Ensure every code block has a language specifier. For code tabs, use the exact Docusaurus `Tabs`/`TabItem` import pattern from `references/tutorial-template.md`.
+Ensure every code block has a language specifier. For code tabs, use the exact Docusaurus `Tabs`/`TabItem` import pattern from `.agents/skills/tutorial/references/tutorial-template.md`.
 
 ### Tutorial doesn't appear in sidebar