Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 2 additions & 0 deletions CHANGELOG.md
Original file line number Diff line number Diff line change
Expand Up @@ -2,6 +2,8 @@

## Unreleased

- Accept the reviewed Skills `v0.2.0` update from `7483dd4dfd8502fd2db376c13e44fa0cbc464161` to `06365b10b42d1c56308b946187d51ccf1efa7d33` as `0.1.0-canary.9` through a separate protected PR, because the range changes sensitive paths. Adds the `init-project` workflow (replacing `build-project`), defaults packages to the newest stable release behind the 7-day age gate, defers fast-changing tool detail to official docs, and adds machine security and privacy workflows. The automatic sensitivity policy is unchanged.

- Fix the `Privacy Guard` denylist check: validate patterns with the same `git grep -iE` engine that scans them, reject PCRE-only escapes, normalize CRLF and blank private lines, and report an invalid pattern by source and line only. Mirrors vllnt/skills.

- Accept the reviewed Skills update from `352eaf7b167a6761161415711ef9d55db26d91d6` to `7483dd4dfd8502fd2db376c13e44fa0cbc464161` as `0.1.0-canary.8` through a separate protected PR, because the range changes sensitive paths. Adds the `@vllnt/convex-api-keys`, `@vllnt/logger` and `@vllnt/analytics` defaults, marks `@vllnt/ui-core` and `@vllnt/ui-native` experimental opt-in, and moves `@vllnt/dig` to not yet default. The automatic sensitivity policy is unchanged.
Expand Down
2 changes: 1 addition & 1 deletion VERSION
Original file line number Diff line number Diff line change
@@ -1 +1 @@
0.1.0-canary.8
0.1.0-canary.9
2 changes: 1 addition & 1 deletion plugins/claude/.claude-plugin/plugin.json
Original file line number Diff line number Diff line change
Expand Up @@ -14,5 +14,5 @@
"license": "MIT",
"name": "vstack",
"repository": "https://github.com/vllnt/stack",
"version": "0.1.0-canary.8"
"version": "0.1.0-canary.9"
}
150 changes: 85 additions & 65 deletions plugins/claude/SOURCE.json

Large diffs are not rendered by default.

Original file line number Diff line number Diff line change
Expand Up @@ -6,22 +6,11 @@ How to audit any landing page and return prioritized, concrete fixes. Primary in

## URL Audit Workflow

Use an installed browser tool for rendered/interactive evidence; the optional CLI examples below require matching installed help. If no browser works, continue static content/structure analysis and explicitly mark live layout, console, interaction and speed unknown. Do not demand one tool or invent an audit. Capture supported viewports, not a universal fixed device set.
Use an installed browser tool for rendered/interactive evidence; take its commands from its current help. If no browser works, continue static content/structure analysis and explicitly mark live layout, console, interaction and speed unknown. Do not demand one tool or invent an audit. Capture supported viewports, not a universal fixed device set.

```bash
# 1. Load + structure/copy extraction
agent-browser open <url>
agent-browser snapshot -i # accessibility tree = section order, headings, CTAs, forms

# 2. Visual hierarchy at the 3 breakpoints that matter
agent-browser set viewport 1920 1080 && agent-browser screenshot lp-desktop.png
agent-browser set viewport 768 1024 && agent-browser screenshot lp-tablet.png
agent-browser set viewport 375 667 && agent-browser screenshot lp-mobile.png

# 3. Is the page even healthy? (broken page = lost conversions)
agent-browser console --level error
agent-browser network --failed
```
1. Load the page and capture its accessibility tree: section order, headings, CTAs, forms.
2. Screenshot desktop, tablet, and mobile viewports to judge visual hierarchy.
3. Check health: console errors and failed network requests (a broken page loses conversions).

Read-only. **Never submit forms or trigger a purchase during an audit.**

Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -17,7 +17,7 @@
1. Reuse the caller's validation contract. Record candidate, mode, required criteria, evidence or checks, unacceptable outcomes, and useful review perspectives from the request, applicable instructions, principles, and owning workflow.
2. Have the producer self-assess each criterion as pass, fail, or unverified with evidence. In authorized code-change mode, require a reduction pass before each commit or final uncommitted handoff: compare the complete task delta with its verified baseline, including earlier task commits, staged, unstaged, and new files. Challenge additions with deletion, reuse, consolidation, or a narrower change at the existing owner. Have the caller apply supported reductions, not merely recommend them; justify retained complexity against the current contract and maintainability. Preserve required features, compatibility, security, failure handling, meaningful test detection, and unrelated or mixed-owner edits. Uncertain attribution or necessity is a gap, not permission to delete. Fewer lines alone, minification, hidden work, and deletion quotas are not success. Rerun invalidated checks and renew affected review after reductions; reuse current proof rather than repeat unchanged passes. Read-only work assesses or proposes, never performs code cleanup. Return known defects to the caller before independent review. A score may summarize a defined rubric but cannot replace required proof.
3. For substantive work, obtain review of correctness, completeness, and preserved behavior, and critique of assumptions, necessity, simplicity, and alternatives from reviewers independent of the producer. One reviewer may cover both; add reviewers only for required separation or non-overlapping parallel scopes. Supply scoped criteria, the candidate or diff, and current evidence. Assign checks within each reviewer's tools and authority; return missing required evidence to the caller. Use direct checks for mechanical work. Without subagents, assess sequentially and report missing independence; return `BLOCKED` if required independent review is unavailable.
4. Collect verdicts and findings with criterion, location or scenario, evidence, impact, smallest remedy, and acceptance check. Deduplicate them into required fixes, selected improvements, optional suggestions, and justified rejections. Resolve disagreements with evidence; a disputed required criterion remains open.
4. Collect verdicts and findings with criterion, location or scenario, evidence, impact, smallest remedy, and acceptance check. Before dependent repairs, tie each blocker to the current authorized criteria or a demonstrated regression or safety obligation, including necessary supporting behavior absent from the original issue. Separate evidence of a defect from necessity and authority for its proposed remedy. Honor accepted recovery paths; resolve conflicting requirements rather than silently strengthening or weakening them. Deduplicate into required fixes, selected improvements, optional suggestions, and justified rejections. Resolve disagreements with evidence; a disputed required criterion remains open.
5. Return fixes to the caller. After authorized revisions, send affected reviewers the reason, revised material, and invalidated evidence. Rerun invalidated checks and renew affected verdicts; retain only evidence and acceptance unaffected by changes to the candidate, inputs, or relevant environment.
6. Return `PASS` only when every required criterion has proof, required perspectives accept their scopes, and no required finding remains. Otherwise return `REVISE` with fixes for the caller to implement and resubmit, or `BLOCKED` with the unmet criterion, cause, options, recommendation, and consequence. On unchanged evidence, change approach rather than retrying or lowering criteria.

Expand Down
23 changes: 0 additions & 23 deletions plugins/claude/skills/build-project/SKILL.md

This file was deleted.

This file was deleted.

Loading
Loading