Required protocol for all agents — status reporting, Gitea API, transactions, and tooling conventions
How this skill is triggered — by the user, by Claude, or both
Slash command
/wasteland-orchestrator:agent-protocolThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
**This is mandatory for all agents.** Hooks will verify compliance.
This is mandatory for all agents. Hooks will verify compliance.
Every agent MUST report status. The dashboard and future visual HQ depend on this.
source ~/.claude/lib/agent-status.sh
export CLAUDE_AGENT_NAME="your-agent-name" # e.g., "pm", "dev-lead-ms", "pixel"
agent_status_update "idle" "Ready"
agent_status_update "working" "Description of current task" "tquick/repo-name" 42
| State | When | Visual (future HQ) |
|---|---|---|
idle | No active task | Agent in lounge |
working | Actively on a task | Agent at desk, typing |
reviewing | Reviewing code/PR | Agent at review station |
brainstorming | Design/brainstorm session | Agent at whiteboard |
meeting | Meeting scribe is live | Agent in meeting room |
blocked | Waiting on dependency/human | Agent pacing |
starting | Session initializing | Agent walking in |
stopping | Session ending | Agent walking out |
Call agent_status_update again with new state/task. Heartbeat is automatic via hooks.
agent_status_clear
NEVER use raw curl to Gitea. Always use the shared library.
source ~/.claude/lib/gitea-api.sh
# GET request
gitea_get "repos/tquick/meeting-scribe/issues" "state=open&limit=10"
# POST request (simple JSON)
gitea_post "repos/tquick/meeting-scribe/issues/1/comments" '{"body":"Comment text"}'
# PATCH request
gitea_patch "repos/tquick/meeting-scribe/issues/1" '{"state":"closed"}'
# Create issue (handles complex bodies safely)
gitea_create_issue "tquick/meeting-scribe" "Issue title" "Issue body" "1,2,3"
git.wastelandwares.com) may change — library reads from envlocalhost:3003 — BLOCKED BY HOOK. Use git.wastelandwares.com.project-management.wastelandwares.com — Being repurposed for pipeline monitoring.All dev work MUST happen in worktrees:
# Create worktree for a story
git worktree add ../.worktrees/issue-17-assistant-skeleton -b feat/issue-17-assistant-skeleton
# Work in the worktree
cd ../.worktrees/issue-17-assistant-skeleton
# Clean up after merge
git worktree remove ../.worktrees/issue-17-assistant-skeleton
Hook enforces: Dev agents get warnings if they cd into the main working tree.
Every meaningful unit of work should be wrapped in a transaction.
source ~/.claude/lib/agent-tx.sh
# Start a transaction — state your intent and justify it
tx_begin "Implementing rolling summary" "Sprint 1 story #18" "tquick/meeting-scribe" 18
# Log each significant action with what + why
tx_action "Created src/summary_prompt.py" "Prompt template for condensed meeting minutes"
tx_action "Updated pipeline.py" "Integrated summary into batch loop"
# End the transaction with outcome
tx_end "success" "Rolling summary working, broadcasts every 2 minutes"
success — completed as intendedpartial — some goals met, others deferredfailed — couldn't complete, needs retry or redesigncancelled — abandoned intentionallytx_action "Added Ollama health check" "Startup should fail gracefully if Ollama is down"tx_action "Edited main.py line 42" "Changed code"Every agent MUST update their persona file (~/.claude/agents/CLAUDE.{name}.md) with learnings after each session.
feat: new featurefix: bug fixdocs: documentationinfra: infrastructure/toolingrefactor: restructuringtest: test changeschore: maintenance# At session start:
source ~/.claude/lib/agent-status.sh
source ~/.claude/lib/gitea-api.sh
source ~/.claude/lib/agent-tx.sh
export CLAUDE_AGENT_NAME="my-name"
agent_status_update "idle" "Ready"
# Begin focused work:
tx_begin "What I'm doing" "Why I'm doing it" "tquick/repo" 42
tx_action "What changed" "Why it changed"
tx_end "success" "Brief summary of outcome"
# Gitea operations:
issues=$(gitea_get "repos/tquick/meeting-scribe/issues" "state=open")
gitea_post "repos/tquick/meeting-scribe/issues/1/comments" '{"body":"Done!"}'
# At session end:
agent_status_clear
npx claudepluginhub wastelandwares/wasteland-orchestratorWraps coding sessions with multi-agent analysis: git status/diff check, CLAUDE.md/context.md updates, automation scouting, learning extraction, follow-up suggestions, duplicate validation, and user-directed action execution.
Guides programmatic control of Claude Code sessions via Claude Agent SDK in TypeScript/JavaScript or Python. Supports custom agents, tools, streaming, and event handling for building AI agents.
Creates bite-sized, testable implementation plans from specs or requirements, with file structure and task decomposition. Activates before coding multi-step tasks.