Skip to content

feat(cli): add deployment agent skill for Claude Code#118

Merged
mmurrs merged 3 commits into
masterfrom
feat/deploy-agent-skill
Apr 8, 2026
Merged

feat(cli): add deployment agent skill for Claude Code#118
mmurrs merged 3 commits into
masterfrom
feat/deploy-agent-skill

Conversation

@mmurrs
Copy link
Copy Markdown
Contributor

@mmurrs mmurrs commented Apr 3, 2026

Summary

What this enables

An AI agent given this skill can autonomously:

  1. Check prerequisites (auth, billing, Docker, environment)
  2. Cross-compile and push Docker images for TDX TEEs
  3. Deploy via pre-built image or verifiable build pipeline
  4. Poll for running status and health check the app
  5. Upgrade, rollback (via release history), or terminate
  6. Parse CLI output for error recovery (no --json on most commands)

Key design decisions

  • Single file: one SKILL.md (~360 lines) — agents use ecloud <command> --help for exact flags rather than a separate reference doc
  • Gate-based structure: sequential gates with explicit pass/fail checks and fix commands — agents stop on failure, fix, retry
  • Output parsing over exit codes: the CLI returns exit 1 for all errors, so the skill teaches agents to parse stdout/stderr content
  • Honest about unknowns: instance type suffix convention (t/s/v) is marked as inferred, pricing is marked as approximate, only TDX is confirmed in official docs

Test plan

🤖 Generated with Claude Code

mmurrs and others added 3 commits April 3, 2026 15:59
Adds a structured SKILL.md runbook that enables AI agents (Claude Code,
etc.) to deploy, upgrade, verify, and manage EigenCloud apps via the
ecloud CLI. Built from live CLI discovery (v0.4.0) and verified against
current docs.

Includes:
- Sequential gate-based deployment flow with pass/fail checks
- Verifiable build pipeline instructions
- Instance type reference with TEE mapping
- Error recovery table with output parsing
- Rollback procedure
- Full CLI flag reference (references/cli-reference.md)

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The CLI's own --help output is the authoritative flag reference.
No need to duplicate it in a separate file.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Additions from end-to-end deploy testing (CitiBike API to EigenCompute):
- Expanded environment docs: sepolia = testnet first, mainnet-alpha = production
- Gate 4: post-deploy profile setup (dashboard naming, verifiable builds)
- Note that --name flag != dashboard display name (requires profile set)
- Profile name cannot contain spaces
- 4 new error recovery entries (425 logs, unnamed apps, profile names, mainnet delay)
- 3 new internal notes (attestation not surfaced, mainnet slower, tee-0x hostname)

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
@mmurrs mmurrs merged commit b7bc75a into master Apr 8, 2026
3 checks passed
@mmurrs mmurrs deleted the feat/deploy-agent-skill branch April 8, 2026 04:11
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants