LACP v0.4.0 is out - added a "thinking partner" layer to the agent...
Nyk 🌱@nyk_builderz
13 views
Mar 28, 2026
~2 min read
Advertisement
1
LACP v0.4.0 is out - added a "thinking partner" layer to the agent control plane.
Most AI workflows optimize for output.
This optimizes for how you think before, during, and after a session:
→ focus brief injected at session start (what problem, what I believe, what's undecided, 30-day goal)
→ thinking nudge detects bare questions and prompts you to state your position first
→ blind spot analysis surfaces unchallenged assumptions before session ends
→ weekly reflection pulls your session history to ask "what pattern keeps recurring?"
→ 3 context modes: thinking-partner, implementation, review each phase maps to a hook event - orient (SessionStart), think (UserPromptSubmit), challenge (Stop), reflect (CLI). fully composable.
Also: LACP is now public + installable via homebrew.
brew tap 0xNyk/lacp && brew install lacp
github.com/0xNyk/lacp/rel…
x.com/nyk_builderz/s…
Most AI workflows optimize for output.
This optimizes for how you think before, during, and after a session:
→ focus brief injected at session start (what problem, what I believe, what's undecided, 30-day goal)
→ thinking nudge detects bare questions and prompts you to state your position first
→ blind spot analysis surfaces unchallenged assumptions before session ends
→ weekly reflection pulls your session history to ask "what pattern keeps recurring?"
→ 3 context modes: thinking-partner, implementation, review each phase maps to a hook event - orient (SessionStart), think (UserPromptSubmit), challenge (Stop), reflect (CLI). fully composable.
Also: LACP is now public + installable via homebrew.
brew tap 0xNyk/lacp && brew install lacp
github.com/0xNyk/lacp/rel…
x.com/nyk_builderz/s…
2
Session Start (automatic)
When you open Claude Code in any project, session_start.py fires and:
1. Injects your focus brieffrom~/.lacp/focus.mdinto the system context — so Claude knows what you're working on, what you believe, and what decisions are open
2. Loads the context mode if LACP_CONTEXT_MODE=thinking-partner is set in your .env - this tells Claude to challenge your thinking, seek counterarguments, and end exchanges with "what are you not seeing?"
3. Warns if your focus brief is stale
When you open Claude Code in any project, session_start.py fires and:
1. Injects your focus brieffrom~/.lacp/focus.mdinto the system context — so Claude knows what you're working on, what you believe, and what decisions are open
2. Loads the context mode if LACP_CONTEXT_MODE=thinking-partner is set in your .env - this tells Claude to challenge your thinking, seek counterarguments, and end exchanges with "what are you not seeing?"
3. Warns if your focus brief is stale
3
During the Session
- Thinking nudge (thinking_nudge.py) fires on UserPromptSubmit — if you ask a bare question like "What should I do about X?" without stating your position, it nudges you to state your thinking first.
Has a cooldown so it doesn't nag every prompt.
- This only activates if LACP_THINKING_NUDGE=1 or
LACP_CONTEXT_MODE=thinking-partner
- Thinking nudge (thinking_nudge.py) fires on UserPromptSubmit — if you ask a bare question like "What should I do about X?" without stating your position, it nudges you to state your thinking first.
Has a cooldown so it doesn't nag every prompt.
- This only activates if LACP_THINKING_NUDGE=1 or
LACP_CONTEXT_MODE=thinking-partner
4
Session End (automatic)
When Claude tries to stop, stop_quality_gate.py fires:
1.Quality gate— existing rationalization detection
2.Blind spot analysis— if LACP_BLIND_SPOT_ENABLED=1, asks "what assumptions weren't challenged?" and surfaces them as a system message before the session ends
When Claude tries to stop, stop_quality_gate.py fires:
1.Quality gate— existing rationalization detection
2.Blind spot analysis— if LACP_BLIND_SPOT_ENABLED=1, asks "what assumptions weren't challenged?" and surfaces them as a system message before the session ends
5
Weekly (manual)
Run lacp-reflect to review your week - it pulls your provenance chain, shows session counts per project, and generates reflection prompts like "What did I defer or avoid?
Run lacp-reflect to review your week - it pulls your provenance chain, shows session counts per project, and generates reflection prompts like "What did I defer or avoid?
