v0.8.0 — agentic CLI guide compliance

Aligns framlit-cli with the agentic CLI guide so coding agents (Claude Code, Cursor, Codex, Devin) can use it productively without burning context.

Highlights

• Universal --json payloads on narration generate and campaign plan (was only on brand / batch / variations). Variants: --json '<inline>', --json - (stdin), --json-file <path>. Validation runs on the resolved payload regardless of input source.
• Client-side --fields projection — e.g. --fields "data.projectId,projects[].name". Shrinks bulky list/get responses so agents don't blow up their context window. Supports nested + array masks.
• Opt-in --sanitize strips known prompt-injection markers (Ignore previous instructions, role markers, </system> tags) from text inputs (brief, prompt, instruction) before dispatching. Stripped lines reported on stderr.
• MCP service filter: framlit mcp --services narration,campaign,brand (or FRAMLIT_MCP_SERVICES= env) subsets the 29-tool registry down to a category list. Filtered tools are also hidden from CallToolRequest, not just listing.
• Agent docs at repo root:
  • AGENTS.md — README-for-AI entry point (setup, invariants, exit codes, costs)
  • SKILL.md — Anthropic Agent Skills format with rules + happy paths for each command
  • CONTEXT.md — vocabulary, services map, security model, schema introspection
  • All three ship with the npm tarball.

Stats

• 87/87 tests pass (was 72 — added 15 for field-mask + sanitizeUntrustedText)
• 117 files / 94 KB tarball (no new runtime deps)
• No breaking changes — all v0.7.x flags still work

Install

npm install -g framlit@0.8.0

🤖 Generated with Claude Code