Promote interceptor skill from repo-local to ~/.pai/skills/ with PAI-flavoured triggers

[virtualian]

Context

A repo-local interceptor skill exists at ~/projects/interceptor/.agents/skills/interceptor/ (with SKILL.md, agents/, references/). It loads automatically when the working directory is the interceptor repo, via the .agents/skills/ ↔ .claude/skills/ symlink convention documented in Interceptor's README §"Agent Instructions".

It is not installed at ~/.pai/skills/interceptor, so it does not appear in the global PAI skill set when working in any other repo. PAI's skill discovery scans ~/.pai/skills/ (and ~/.claude/skills/ for Claude Code), not arbitrary repo-local paths — so this is a distribution gap, not an authoring gap.

Current state

~/projects/interceptor/.agents/skills/interceptor/
├── SKILL.md
├── agents/
└── references/

The skill's frontmatter description currently reads:

"Use when Codex should use the interceptor CLI to operate a live signed-in browser or macOS app…"

That framing is Codex-specific. PAI skills follow a different trigger convention (e.g. Scraping/SKILL.md):

"…USE WHEN scraping, Bright Data, proxy, crawl, scrape URL, Twitter scraping, …"

Gap

1. Distribution — interceptor skill is invisible to PAI outside the interceptor repo.
2. Trigger surface — Codex-flavoured description does not match PAI's USE WHEN <verb-noun>, … convention, so PAI's skill matcher will not surface it on natural language requests like "open this URL in my logged-in browser" or "extract LinkedIn event attendees".
3. Naming — repo-local skill is lowercase interceptor; PAI skills tree uses TitleCase (Research, Scraping, Investigation, …). Decide whether to rename or leave lowercase deliberately.

Proposed work

• Stage a PAI-flavoured copy of the skill into ~/.pai/skills/Interceptor/ (or interceptor/ — TBD).
• Rewrite the frontmatter description with PAI-style USE WHEN <trigger>, <trigger>, … triggers covering the skill's real surface area: browser automation, signed-in session, scrape live page, extract LinkedIn event, click element, fill form, read network traffic, drive Canva / Google Docs / Slides, native macOS automation, AX tree, record session, replay session.
• Decide whether to symlink to the canonical repo location (~/projects/interceptor/.agents/skills/interceptor/) or stage an independent copy. Symlink keeps the source of truth in the interceptor repo; copy decouples PAI from a project repo's lifecycle.
• If symlinking, document this as a precedent for "skill lives in its product repo, PAI links into it" — useful pattern for future product-bound skills.
• Verify skill loads in a fresh PAI session in an unrelated CWD.

Related

• Skills separation is documentation-only: Claude Code harness reads ~/.claude/skills/, not ~/.pai/skills/ #110 — Skills separation is documentation-only (~/.claude/skills/ vs ~/.pai/skills/ resolution gap).
• Resync Research and Learning pack sources into ~/.pai/skills/ runtime #145 — Resync Research and Learning pack sources into ~/.pai/skills/ runtime.
• Pack commands never staged into ~/.pai/commands/ — #113 migration has no source path #146 — Pack commands never staged into ~/.pai/commands/ — adjacent staging-pipeline issue.

Open questions

• Should the canonical location for product-specific skills be the product repo (with PAI symlinking in), or ~/.pai/skills/ (with the product repo symlinking out)? This issue is one instance of the wider question.
• Naming: keep lowercase interceptor (matches repo + binary) or TitleCase Interceptor (matches PAI convention)?