@arthai/agents 1.0.0

package/skills/share/SKILL.md

@@ -0,0 +1,211 @@
+---
+name: share
+description: "Format plans and strategies for sharing. Adds attribution, cleans sensitive info, outputs shareable format. Usage: /share <plan-name> <--format md|slack|twitter>"
+user-invocable: true
+arguments: "<plan-name> <--format md|slack|twitter>"
+---
+
+# Share Skill — Shareable Artifact Generator
+
+Formats generated plans, strategies, and briefs for sharing outside the terminal.
+Enables the viral artifact loop: user generates a deliverable → shares it → peers see
+quality → peers try the toolkit.
+
+**Design principles:**
+- The output IS the marketing — make it look professional
+- Attribution is subtle, not annoying
+- Strip sensitive info by default (API keys, internal URLs, file paths)
+- Multiple output formats for different sharing contexts
+
+## Argument Parsing
+
+Parse from the argument string:
+- `plan-name`: Optional. Name of a plan in `.claude/plans/`. If omitted, share the most recently generated artifact from the current session.
+- `--format`: Optional. Output format: `md` (default), `slack`, `twitter`, `html`
+- `--no-attribution`: Optional flag. Skip the attribution footer.
+
+## Execution Steps
+
+### 1. Find the Artifact to Share
+
+**Priority order:**
+1. If `plan-name` is provided, look for `.claude/plans/{plan-name}.md`
+2. If no name, check if a plan was generated in this session (look for recent Write tool calls to `.claude/plans/`)
+3. If still nothing, list available plans and ask the user to pick:
+   ```
+   Available plans:
+     1. user-notifications.md (created 2 hours ago)
+     2. oauth-support.md (created yesterday)
+     3. gtm-strategy.md (created 3 days ago)
+   Which one would you like to share?
+   ```
+
+### 2. Read and Process the Artifact
+
+Read the plan file. Then process it:
+
+**Strip YAML frontmatter:**
+Remove the `---` delimited YAML block at the top (contains internal metadata like `layers: [frontend, backend]`).
+
+**Strip sensitive information:**
+- Remove absolute file paths (replace `/home/user/project/src/...` with `src/...`)
+- Remove API keys, tokens, secrets (anything matching common patterns)
+- Remove internal URLs (localhost, staging URLs, internal domains)
+- Remove `.claude/` references and toolkit internals
+
+**Clean up for readability:**
+- Ensure headings are properly formatted
+- Remove any `<!-- comments -->`
+- Ensure code blocks have language tags
+
+### 3. Format for Target Platform
+
+#### Markdown (default — `--format md`)
+
+Output the cleaned plan as a well-formatted markdown document with:
+- A title block (feature name, date)
+- All sections preserved
+- Attribution footer
+
+```markdown
+# Feature Plan: User Notifications
+
+*Generated on March 20, 2026*
+
+## Problem & User Segment
+[content...]
+
+## Technical Approach
+[content...]
+
+## Task Breakdown
+[content...]
+
+## Acceptance Criteria
+[content...]
+
+---
+*Created with [Claude Agents](https://github.com/ArthTech-AI/claude-agents) — AI product + architecture team*
+```
+
+#### Slack (`--format slack`)
+
+Format for pasting into Slack with Slack-compatible markdown:
+- Use `*bold*` instead of `**bold**`
+- Use bullet points that render well in Slack
+- Shorter sections (Slack truncates long messages)
+- Attribution as a small footer line
+
+```
+*Feature Plan: User Notifications*
+
+*Problem & User Segment*
+• [concise bullet points...]
+
+*Technical Approach*
+• [concise bullet points...]
+
+*Key Tasks*
+• [top 5 tasks only...]
+
+_Created with Claude Agents — AI product + architecture team_
+```
+
+#### Twitter/X (`--format twitter`)
+
+Format as a thread-ready summary (each section = one tweet-length block):
+- Max 280 chars per block
+- Use line breaks between blocks
+- Include a hook/opener
+- Attribution in the last block
+
+```
+Just used AI agents to plan our notifications feature in 8 minutes.
+
+Here's what came out 👇
+
+---
+
+Problem: Users miss critical updates because email-only notifications
+have <12% open rates. Mobile push + in-app needed.
+
+Target: Active users with >3 logins/week (power users first).
+
+---
+
+Technical approach:
+- WebSocket for real-time in-app
+- FCM/APNs for push
+- Preference center per channel
+- Background job queue for batching
+
+---
+
+5 backend tasks, 4 frontend tasks, 8 acceptance criteria.
+Full plan generated by an AI PM + Architect team.
+
+Built with Claude Agents → github.com/ArthTech-AI/claude-agents
+```
+
+#### HTML (`--format html`)
+
+Wrap the markdown in a clean, minimal HTML template suitable for sharing as a link:
+- Professional styling (system fonts, good spacing, readable line length)
+- Responsive (mobile-friendly)
+- Print-friendly
+- Meta tags for social sharing (og:title, og:description)
+
+### 4. Output the Result
+
+**For md/slack/twitter:** Write the formatted output directly to the terminal for the user to copy.
+
+**For html:** Write to `.claude/shared/{plan-name}.html` and inform the user:
+```
+Shareable version saved to .claude/shared/{plan-name}.html
+Open it in a browser to preview, or share the file with your team.
+```
+
+### 5. Offer Next Steps
+
+After outputting, suggest:
+
+```
+Sharing options:
+  - Copy the text above and paste into Slack/docs/email
+  - "share this as slack" for Slack-formatted version
+  - "share this as twitter" for tweet thread format
+  - "share this as html" for a shareable web page
+```
+
+## Attribution Footer
+
+Default attribution (included unless `--no-attribution`):
+
+For plans with multiple agents:
+```
+Created with Claude Agents — PM + Architect team
+```
+
+For GTM strategies:
+```
+Created with Claude Agents — GTM Expert
+```
+
+For design briefs:
+```
+Created with Claude Agents — Design Studio
+```
+
+The attribution is intentionally subtle — one line, no logos, no CTAs.
+It works because people will ask "what's Claude Agents?" when they see
+a high-quality deliverable with that attribution.
+
+## Rules
+
+- **Never share raw file paths** — always strip to relative paths
+- **Never share API keys or secrets** — scan and remove before outputting
+- **Keep attribution subtle** — one line, not a banner
+- **Preserve the substance** — don't summarize or truncate unless format requires it (twitter)
+- **Default to markdown** — it's the most portable format
+- **Don't auto-share** — always output to the terminal for the user to copy; never post anywhere
+- **Respect --no-attribution** — if the user opts out, honor it completely