Skip to content

refresh_agent_readme_survey.md

Latest commit

 

History

History
94 lines (73 loc) · 8.13 KB

refresh_agent_readme_survey.md

File metadata and controls

94 lines (73 loc) · 8.13 KB
Raw

Agent Landscape Survey

DeepWiki- and upstream-backed review for TeaAgent competitive positioning. TeaAgent stays a governance-first harness; this artifact tracks parity, gaps, and backlog actions — not a second agent framework.

Last reviewed: 2026-06-06

Related evidence:

Sources reviewed

Project DeepWiki / upstream URL README URL Reviewed signals TeaAgent parity Gap / differentiator Backlog action Local doc ref
OpenAI Codex https://deepwiki.com/openai/codex https://github.com/openai/codex Multi execution modes, sandboxing, MCP, IDE, cloud surfaces Tool registry, MCP HTTP/stdio, Code Mode sandbox profiles, VS Code extension, USAGE mode/safety matrix Hosted/cloud surface docs remain thin P2 maintenance docs/architecture.md, docs/use-cases.md
Claude Code https://deepwiki.com/anthropics/claude-code https://github.com/anthropics/claude-code Subagents, hooks, MCP, background sessions, permission modes, managed settings subagent/subagent_batch with lineage + shared/worktree/container isolation, hooks, permission modes, MCP, skills Background session cloud docs remain thin P2 maintenance docs/architecture.md, docs/use-cases.md
OpenCode https://deepwiki.com/sst/opencode https://github.com/anomalyco/opencode Provider breadth, client-server, plugins, skills, MCP, desktop, VS Code 13 providers, plugins/skills, MCP, ACP/VS Code, USAGE surface recipes Client-server/desktop hosted docs remain thin P2 maintenance docs/architecture.md, docs/use-cases.md
Hermes https://deepwiki.com/NousResearch/hermes-agent https://github.com/nousresearch/hermes-agent Persistent digital employee, long-term memory loop, agent-created skills, multi-platform messaging gateway, sandbox backends, cron scheduling Memory curation, skills/plugins, subagent isolation, Plan/Auto modes, CI workflows Cross-platform resident entry, self-generated skills, persistent automation use case not yet covered P2 maintenance docs/architecture.md, docs/use-cases.md
OpenHands https://deepwiki.com/OpenHands/OpenHands https://github.com/All-Hands-AI/OpenHands SDK/CLI/GUI/cloud/enterprise, sandbox-decoupled V1 Managed runtime stubs, MCP, audit, Code Mode, use-case dashboard Hosted/cloud surface docs are stub-level only P2 maintenance docs/use-cases.md
Aider https://deepwiki.com/Aider-AI/aider https://github.com/Aider-AI/aider Repo-map context, edit strategies, git workflow Workspace tools, LSP/code-analysis, GraphQLite, preflight context_pack with LSP + hybrid/knowledge/GraphQLite read-only hits Whole-repo map heuristics still thinner than Aider's dedicated repo-map UX P2 maintenance docs/use-cases.md
LangGraph https://deepwiki.com/langchain-ai/langgraph https://github.com/langchain-ai/langgraph Graph state, checkpoints, durable execution CheckpointStore, runner limits, audit chain No graph-native orchestration (intentional harness boundary) Document as non-goal docs/use-cases.md
CrewAI https://deepwiki.com/crewAIInc/crewAI https://github.com/crewAIInc/crewAI Role-based crews, task delegation A2A delegation, ANP governed federation No multi-role crew DSL (intentional harness boundary) Document as non-goal docs/use-cases.md
GitHub Copilot cloud agent https://docs.github.com/en/copilot/concepts/agents/cloud-agent/about-cloud-agent https://docs.github.com/en/copilot/how-tos/use-copilot-agents/cloud-agent/start-copilot-sessions PR-native async background work, GitHub Actions environments, multi-entry intake Audit, policy, MCP, automation hooks No GitHub-native cloud agent surface P2 adapter research docs/analysis/competitor-self-comparison-matrix-2026-06-06.md
Cursor https://docs.cursor.com/background-agent https://docs.cursor.com/agent IDE agent modes, background remote agents, explicit CLI permissions CLI/TUI governance, audit chain, local-first No productized remote background agent P2 UX benchmark only docs/analysis/competitor-self-comparison-matrix-2026-06-06.md
Cline https://docs.cline.bot/cline-overview https://docs.cline.bot/core-workflows/plan-and-act Plan/Act separation, per-action approval, IDE + terminal Plan-before-write gates, approval manager, run evidence Plan/Act UX less legible in TeaAgent surfaces P1 conversation trust (WS1) docs/analysis/competitor-self-comparison-matrix-2026-06-06.md

Additional README baselines (prior survey): Claude Code GitHub, Codex CLI GitHub, Cursor docs, Gemini CLI, Continue — still valid for env/skill conventions.

June 2026 refresh notes

Official docs checked on 2026-06-06 confirm three strategic pressures:

  1. Remote async agents are baseline for major vendors (Codex, Copilot, Cursor, Kiro, Devin, Jules).
  2. IDE-native and PR-native entry beats terminal-only onboarding for daily use.
  3. Audit/compliance surfaces (Codex Compliance API, Devin enterprise logs) are becoming purchase criteria — TeaAgent should lead with provable audit once durability gates pass.

Full row-level comparison: docs/analysis/competitor-self-comparison-matrix-2026-06-06.md.

Implemented parity (harness core)

  • Tool registry with schema validation and destructive approval
  • Audit chain with redaction
  • MCP stdio + streamable HTTP
  • A2A discovery/delegation
  • ACP IDE adapter
  • ANP governed federation boundary (ANPGovernedService)
  • OAuth refresh-token rotation (ADR 0004)
  • Google managed runtime (GoogleADKRuntime, VertexAgentRuntime)
  • Permission modes, Plan Mode, Auto Mode, Code Mode
  • Provider/docs consistency acceptance (test_provider_matrix_consistency_flow.py)
  • Preflight read-only context_pack (hybrid, .teaagent/knowledge, GraphQLite DB hits; test_context_pack_read_only_flow.py)
  • Subagent delegation with shared / worktree / container isolation and lineage audit (test_subagent_*_isolation_flow.py)
  • Everyday usage docs: agent daily start ritual, task recipes, context profiles, TUI loop (README + USAGE + cli)

Competitive differentiators (implemented / maintenance)

Track Why it matters Backlog ID
Docs/provider drift guard README/USAGE/architecture/runtime provider registry aligned; shared credential env vars handled Done
Subagent lineage + isolation Parent run id, depth, batch index; isolation: shared, worktree, or container snapshot Done (maintenance)
Repo-map / context pack Preflight context_pack surfaces candidate files/memories without writes Done (maintenance)
Mode and safety matrix Permission modes, Plan/Auto/Code lanes, approvals, rollback Done (maintenance)
Multi-surface recipes CLI, TUI, VS Code, MCP, ACP, A2A, ANP, managed runtime Done (maintenance)
Plugin/skill catalog docs/plugin-skill-catalog.md + fixtures Done (maintenance)
Use-case dashboard refresh Matrix/HTML include survey date and open-gap counts Done (maintenance)
Recurring survey cadence docs/release-checklist.md Done (maintenance)

Current API note (non-breaking)

subagent / subagent_batch record parent-child run lineage and accept isolation = shared | worktree | container (filesystem snapshot under .teaagent/subagent-containers/). Keep these fields additive-compatible for future subagent tooling.

Future research / P2 maintenance

  • Hosted/cloud surface docs for managed-runtime deployments.
  • Background-session story across CLI/TUI/IDE surfaces.
  • Desktop/client-server packaging guidance.
  • Repo-map quality evaluation against larger real-world repositories.

Next review trigger

Re-run this survey before the next minor release or when adding a new federation/protocol ADR. Update docs/backlog-priority.md and docs/use-cases.md differentiator sections when signals change.