Claude Code skills — Markdown SOPs + external
integrations auto-loaded at SessionStart from
~/.claude/skills/ and ./.claude/skills/.
Toggle writes to ~/.concinno/skills.json as the
source-of-truth; a future SessionStart hook will honour it
(currently advisory). Includes MCP / Agent / Hammer and any other
skill package.
Browse + install + uninstall first-party
concinno-skills-* sub-packages from PyPI. The
installed list also surfaces hook-only sub-pkgs
that don't ship a SKILL.md (so a fresh
pip install concinno-skills-memory shows up
here even when the Skills tab can't see it). Twice-click confirm
gates pip subprocess unless
release_authorization.disabled=true.
Installed
Available
Slash commands the host agent pops up when you type /
(Claude Code reads ~/.claude/commands/; Cursor / Codex
adapters will read their own conventions when shipped).
Concinno-managed commands are tagged
concinno. Click Resync to
reinstall after a Concinno upgrade.
Host-agent bash sandbox permission rules (Claude Code today,
Cursor / Codex tomorrow via portable adapters). This is the
second gate, separate from concinno
release_authorization. Concinno
disabled=True does not open this layer. Edit the JSON
file directly and restart the host agent to pick up changes.
ZIQ auto-tune posterior (~/.concinno/ziq_posterior.json).
Read-only here — actual tuning happens in the online-learning loop.
Manual values on the Features tab always win when explicitly
pinned 🔒. Feature-level ZIQ toggle turns the whole feature opaque
to the tuner.
Live runtime-state snapshot — polled every 3s.