nyldn/claude-octopus v11.15.1 on GitHub

/octo:discover and /octo:research now use the quick, standard, or deep
intensity you selected. The command passes --intensity= to the research
engine, avoiding a bracket token that became prompt text in Bash and failed
as an unmatched glob in zsh. Research focus and output format stay in the
prompt. #1213,
commit fa1eab5.
Knowledge-work and security-audit examples use supported command arguments.
Parallel workflow examples pass the previous phase's result file positionally
so the next phase receives it. Focus, depth, and audit scope go in the prompt.
#1223,
commit 733e717.
SessionEnd writes preferences and Markdown session learnings to the current
project's existing Claude auto-memory directory; SessionStart restores
preferences from that directory. The hooks no longer create repository-local
memory/ or fall back to another project's storage. Custom CLAUDE_CONFIG_DIR
and CWD fallback use the same mapping at session start and end. Structured
cross-task JSON learnings remain shared across projects under
~/.claude-octopus/learnings and are not automatically read by SessionStart.
Fixes #1183.
#1225,
commit 286ef4d.

Add regression coverage for orchestrator EXIT cleanup when ownership is
unknown. #1210,
commit c06ffc0.