Skip to content

parcadei/Continuous-Claude-v3 / .claude/skills/parallel-agents

parallel-agents

Unverified

Parallel Agent Orchestration

Parallel Agent Orchestration

When launching multiple agents in parallel, follow this pattern to avoid context bloat.

Core Principles

  1. No TaskOutput calls - TaskOutput returns full agent output, bloating context
  2. Run in background - Always use run_in_background: true
  3. File-based confirmation - Agents write status to files, not return values
  4. Append, don't overwrite - Multiple agents can write to same status file

Output Patterns

Simple Confirmation (parallel batch work)

For tasks where agents just need to confirm completion:

# Agent writes to shared status file
echo "COMPLETE: <task-name> - $(date)" >> .claude/cache/<batch-name>-status.txt
  • Use >> to append (not > which overwrites)
  • Include timestamp for ordering
  • One line per agent completion
  • Check with: cat .claude/cache/<batch-name>-status.txt

Detailed Output (research/exploration)

For tasks requiring detailed findings:

.claude/cache/agents/<task-type>/<agent-id>/
├── output.md      # Main findings
├── artifacts/     # Any generated files
└── status.txt     # Completion confirmation
  • Each agent gets own directory
  • Full output preserved for later reading
  • Status file still used for quick completion check

Task Prompt Template

# Task: <TASK_NAME>

## Your Mission
<clear objective>

## Output
When done, write confirmation:
\`\`\`bash
echo "COMPLETE: <identifier> - $(date)" >> .claude/cache/<batch>-status.txt
\`\`\`

Do NOT return large output. Complete work silently.

Launching Pattern

// Launch all in single message block (parallel)
Task({
  description: "Task 1",
  prompt: "...",
  subagent_type: "general-purpose",
  run_in_background: true
})
Task({
  description: "Task 2",
  prompt: "...",
  subagent_type: "general-purpose",
  run_in_background: true
})
// ... up to 15 parallel agents

Monitoring

# Check completion status
cat .claude/cache/<batch>-status.txt

# Count completions
wc -l .claude/cache/<batch>-status.txt

# Watch for updates
tail -f .claude/cache/<batch>-status.txt

Batch Size

  • Max 15 agents per parallel batch
  • Wait for batch to complete before launching next
  • Use status file to track which completed

DO

  • Use run_in_background: true always
  • Have agents write to status files
  • Use append (>>) not overwrite (>)
  • Give each agent clear, self-contained instructions
  • Include all context in prompt (agents don't share memory)

DON'T

  • Call TaskOutput (bloats context)
  • Return large outputs from agents
  • Launch more than 15 at once
  • Rely on agent return values for orchestration

Example: Provider Backfill

# Status file
.claude/cache/provider-backfill-status.txt

# Each agent appends on completion
echo "COMPLETE: anthropic - Thu Jan 2 12:34:56 2025" >> .claude/cache/provider-backfill-status.txt
echo "COMPLETE: openai - Thu Jan 2 12:35:12 2025" >> .claude/cache/provider-backfill-status.txt

Check progress:

cat .claude/cache/provider-backfill-status.txt
# COMPLETE: anthropic - Thu Jan 2 12:34:56 2025
# COMPLETE: openai - Thu Jan 2 12:35:12 2025
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 463kUpdated Claude CodeCodexVerifiedInstall
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 463kUpdated Claude CodeCodexVerifiedInstall
discernment-nudgeanthropics/skillsAfter you give a substantive answer or draft that the user may act on — advice or recommendations, drafted artifacts such as goals, plans, pitches, proposals, or emails, estimates or projections, analysis or interpretation of data, factual claims they may rely on, or a multi-step argument — invoke this skill BEFORE finalizing your reply and then, if it applies, append 2-3 short follow-up questions, each tied to something specific in what you just produced, that help the user check key facts, probe the reasoning or assumptions, and notice missing context. Do this at most once per conversation. Skip it when the user asked a trivial how-to or simple lookup, wants a purely educational explanation, asked you only to format, convert, or assemble a file from content they provided, is writing code they will run, is doing creative writing or casual chat, or already asked you to double-check, cite, or review — the skill file explains these boundaries and the exact output format.Stars: 195kUpdated Claude CodeCodexVerifiedInstall
internal-commsanthropics/skillsA set of resources to help me write all kinds of internal communications, using the formats that my company likes to use. Claude should use this skill whenever asked to write some sort of internal communications (status reports, leadership updates, 3P updates, company newsletters, FAQs, incident reports, project updates, etc.).Stars: 195kUpdated Claude CodeCodexVerifiedInstall