sync-agent-skillslisted

# Sync Agent Skills Synchronize from canonical skill source through each agent's supported native format. Folder existence is not proof that a running agent can discover a skill. ## Safety - Identify the canonical source and its revision before touching provider projections. Never edit generated installed copies as source. - Prefer a receipt-backed installer such as Dufflebag. Preserve the existing selected feature set and make additions explicit; do not replace the user's bag with only the newly named skills. - Detect providers from the installer's catalog and live evidence. Never invent a dot-directory or claim support for an uncataloged agent such as Grok without an official, verified integration surface. - Plan before apply. Preserve non-owned provider instructions and config byte-for-byte; only the receipt authorizes updates or deletion. - Do not expose tokens or modify unrelated agent, MCP, model, or permission settings. ## Workflow 1. Inspect repository instructions, canonical skill directories, catalog entries, shipped-path allowlists, current revision, dirty state, and existing global or project receipt. 2. Validate every requested skill before installation: frontmatter name/description, referenced assets, source/catalog identity, and exact shipped paths. 3. Detect installed agents with the repository's supported evidence model. Produce a matrix of provider ID, detection evidence, native target format, destination, and support status. 4. Determine desired state