spike: orchestrator loop M3-M6 exploration#11
Draft
leandronsp wants to merge 16 commits into
Draft
Conversation
…n, and info command
…ment Import orchestration infrastructure: 3 agents (scout, code-reviewer, plan-reviewer), 3 contextual rules (git, testing, elixir), and 2 new skills (learn, debug). Agents are read-only specialists that /dev wires into its phases — scout for exploration, plan-reviewer for stress-testing plans, code-reviewer for self-review after implementation. Renamed /tdd to /dev and wired agent orchestration into its phases. Merged /techdebt into /review as inline audit tables. Added shell script review expertise (POSIX compliance, modularity, quoting) to both the code-reviewer agent and /review skill. Added Shell Style section and .claude/ directory to CLAUDE.md.
Move client API (get_logs, stop, kill, pause, info, etc.) from Mission into Mission.Client. Split bin/overmind (314 lines) into bin/cli/helpers.sh and bin/cli/commands.sh. Fix build_info field name mismatch (activity_timeout → last_activity), cmd_attach error check order, and cmd_run echo → printf.
Establish a balanced documentation convention across the codebase: targeted comments on non-obvious logic (WHY, not WHAT) while keeping self-documenting code uncommented. Add .claude/rules/shell.md for POSIX shell conventions, mix smoke alias for daemon lifecycle checks, and embed the convention in all agents, rules, and skills.
Port.open now strips CLAUDECODE and CLAUDE_CODE_ENTRYPOINT from child processes, allowing Claude CLI to run as a subprocess inside Claude Code sessions. This unblocks mix e2e and future Overmind skills.
Replace MIT with AGPL v3 for open-core model. Add aerial forest cover image (credit: Ian @photoken123 on Unsplash) in multiple sizes. Rewrite README with mycelium network metaphor and updated usage/roadmap.
…kill M2.5 implements the building blocks for agent orchestration: - Exit code storage in ETS so wait can return actual exit codes - Monitor-based wait command (no polling/sleep) with timeout support - Parent hierarchy via --parent flag with validation and name resolution - ps --tree for hierarchy visualization, ps --children for filtering - Depth-first kill --cascade that tears down entire subtrees - Shell CLI: cmd_wait, --parent, --tree, --children, --cascade flags Also fixes :json.encode encoding nil as "nil" string instead of JSON null in pause/wait responses, which broke attach on sessions without --cwd.
Extract ps/tree rendering into Overmind.Formatter (overmind.ex 231→155 lines), fix O(n²) in render_tree, replace nested inline ifs with multi-clause helpers, handle null exit_code in cmd_wait.
…overmind skill
Status command exposes daemon health (uptime, memory, process counts, mission
breakdown by status). Result command extracts structured output from completed
Claude missions. Missions now receive OVERMIND_MISSION_ID and OVERMIND_MISSION_NAME
env vars for self-awareness. API run response returns {id, name} instead of bare id.
The /overmind skill provides a system prompt template that launches a Claude session
as orchestrator — it decomposes tasks, spawns child agents, and runs a reconciliation
loop until all work converges. CLI gains --json flag on run, result command, status,
and live monitor.
…ashboard Orchestrator loop spike that implemented remaining roadmap items: - Blueprint TOML parsing and DAG-based agent spawning (M3) - Full isolation: worktrees, port registry, Docker (M4) - Quests and Rituals with cron scheduler (M3.5) - k9s-inspired TUI with mission panels (M6) - Phoenix LiveView web dashboard (M6) - Resource monitoring (top command, CPU/MEM per mission) - allowed_tools flag for autonomous Claude agents - Akasha shared memory with SQLite (M5)
13 tasks
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Context
Orchestrator loop test where a Claude agent autonomously implemented M3-M6 features. Code is exploratory and untested as a whole — kept as reference for future milestone work.
🤖 Generated with Claude Code