Imported from Nakul-Kumar/AgentDevOps (
AGENTS.md). Install upstream withnpx skills add Nakul-Kumar/AgentDevOps. Copyright stays with the author.
AGENTS.md - AgentDevOps Engineering Team
Authoritative roster of active agents. Definitions live in .claude/agents/.
Stale product-domain agents (market-research, research-team-lead,
strategy, strategy-opus) were removed during the April 2026 cleanup and
are not listed.
Model names below are the configured Claude CLI/runtime aliases for queued agents. OpenAI/Codex models are optional sidecars for manual escalation and review; they are not the default queue runtime.
Implementation Targets (queue_implementation)
These agents accept tasks submitted via the task queue:
| Agent | Model | Purpose |
|---|---|---|
software-dev |
claude-sonnet-4-6 | Code, bugs, refactors, tests. maxTurns 75. |
data-eng |
claude-sonnet-4-6 | Postgres, Redis, schemas, migrations, pipelines. |
ml-ai |
claude-sonnet-4-6 | Routing experiments, embeddings, DSPy, shadow-log analysis, Kaggle jobs. |
devops |
claude-sonnet-4-6 | Docker, CI, SSH/Tailscale, scripts, secrets, infrastructure authoring. |
release-engineer |
claude-sonnet-4-6 | Deploy, restart sequencing, repo-to-system install via scripts/install_system_files.sh, post-merge verify, rollback playbook. |
docs-maintainer |
claude-haiku-4-5-20251001 | README, AGENTS.md, CLAUDE.md, runbooks, handoffs. |
Advisory / Research / Review (not queue targets)
These agents are coordinated directly or through separate task creation.
| Agent | Model | Purpose |
|---|---|---|
tech-research |
claude-sonnet-4-6 | Scoped technical research, >=2 alternatives, trade-offs. team: research |
deep-researcher |
claude-opus-4-7 | Exhaustive multi-hop research for deep/SOTA surveys. Not code. team: research |
code-critic |
claude-haiku-4-5-20251001 | Adversarial review of core/ and agent infra. Read-only. maxTurns 35. |
planner |
claude-haiku-4-5-20251001 | Task-decomposition mode of project-brain (2-6 ordered subtasks, JSON-only IO). Identity merged into project-brain; this entry preserves the IO contract for existing dispatchers. |
skill-builder |
claude-haiku-4-5-20251001 | Installs skills from URLs; opens draft PRs on auto/skill-* branches. team: platform |
project-brain |
claude-opus-4-7 | Steering, plan decomposition, approval-aware coordination, and operator-facing engineering assistance. team: platform |
Team Frontmatter
Agent files use team: to select context briefs:
team: engineering- software-dev, data-eng, ml-ai, devops, code-criticteam: research- tech-research, deep-researcherteam: platform- project-brain, skill-builder
Agents without explicit team: frontmatter (docs-maintainer) do not receive
a context brief injection.
Routing Notes
- Only
software-dev,data-eng,ml-ai,devops,release-engineer, anddocs-maintainerare validqueue_implementationhandoff targets from planner and project-brain. code-critic,tech-research,deep-researcher,planner, andskill-builderare dispatched directly or via separate task creation.project-brainis a steering agent; it delegates to implementers, not itself.- Current production routing is rule-based heuristics. ML/DSPy/Qwen routing experiments are research scaffolding until promoted by measured evidence.
Removed Agents
The following agents were removed in the April 2026 cleanup and must not be re-added without an explicit operator request:
market-research,market-research-fast- product-domain, removedresearch-team-lead- product-domain orchestrator, removedstrategy,strategy-opus- product-domain strategy agents, removed