jb/diagram-design
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix(packaging): make skill references bundle-safe (#125)

Browse Source 
* fix(packaging): make skill references bundle-safe

* test(packaging): normalize scanner paths portably

* fix(packaging): bundle required runtime helpers
This commit is contained in:
Cathryn Lavery
2026-08-19 20:31:56 -07:00
committed by GitHub
parent 0487290d5a
commit b132fa5d7a
8 changed files with 178 additions and 24 deletions
Download Patch File Download Diff File Expand all files Collapse all files
README.md
+1 -1
View File
@@ -449,7 +449,7 @@ behavior, resource caps, named failures, and reference/command wiring.
Label placement is gated geometrically: `python3 scripts/verify-geometry.py --all` fails CI when a label mask overlaps a node declared later in the document, because the node fill would clip the text at render time. `python3 scripts/test-verify-geometry.py` keeps that checker honest in both directions.
Treemaps get a second geometric gate, because their whole claim is that area *is* the encoding: `python3 scripts/verify-treemap.py --all` fails CI when a cell's share of the drawn area doesn't match the value printed inside it, or when a label overruns the cell it names. It measures area error as a *relative* figure — an absolute one passes exactly the small cells most likely to be wrong. `python3 scripts/test-verify-treemap.py` keeps it honest in both directions.
Docs and routing surfaces are themselves gated: `python3 scripts/verify-docs-sync.py` fails CI if the SKILL.md description loses a type's lexical hook, the gallery can't reach a shipped example, the README tree names a file that doesn't exist, a relative `references/*.md` link in SKILL.md is broken, or the Claude/Pi profile surfaces drift from `profiles.md`. `python3 scripts/test-verify-docs-sync.py` exercises those newer checks adversarially. The skill also ships `skills/diagram-design/scripts/self_check.py` — a distilled output checker installed agents can run on their own generated diagrams; `python3 scripts/test-self-check.py` keeps it honest. Settled design decisions (why one pinned controller, why patterns never add types, the autoplay policy, the SKILL.md byte cap, why label placement is verified geometrically, and why client profiles use marker-first resolution) live as short ADRs in `docs/adr/` — read them before relitigating one, add one when you settle a new policy.
Docs and routing surfaces are themselves gated: `python3 scripts/verify-docs-sync.py` fails CI if the SKILL.md description loses a type's lexical hook, the gallery can't reach a shipped example, the README tree names a file that doesn't exist, a relative reference link is broken, a scanner-visible support path is not shipped inside the skill package, or the Claude/Pi profile surfaces drift from `profiles.md`. `python3 scripts/test-verify-docs-sync.py` exercises those newer checks adversarially, including the strict-bundler behavior used by Hermes Agent. The skill also ships `skills/diagram-design/scripts/self_check.py` — a distilled output checker installed agents can run on their own generated diagrams; `python3 scripts/test-self-check.py` keeps it honest. Settled design decisions (why one pinned controller, why patterns never add types, the autoplay policy, the SKILL.md byte cap, why label placement is verified geometrically, and why client profiles use marker-first resolution) live as short ADRs in `docs/adr/` — read them before relitigating one, add one when you settle a new policy.
All pull requests and pushes are automatically validated across Linux, Windows, and macOS runners via GitHub Actions CI (`.github/workflows/ci.yml`).