Added
- Every LLM contact (doer
generate() and planner planner_complete(), across all provider paths — Anthropic native, OpenAI-compatible, claude-cli/codex-cli shell-out, in-process llama.cpp) now prepends a fixed [OPERATING CONSTRAINTS] block to the prompt:
- No thinking — no chain-of-thought; output only the requested result
- No internet — no browsing/fetching/external network
- Tools — only those strictly necessary
- Skills — only those strictly necessary
- Prepended (not appended) so the template's strict
[OUTPUT] block stays last. Folded into the completion cache key. Opt out with SIMPLICIO_NO_LLM_DIRECTIVES=1.
PyPI: https://pypi.org/project/simplicio-cli/0.7.0/
