Skip to content

AgentDoctor 2.1.0

Choose a tag to compare

View all tags
@pranee54 pranee54 released this 23 Sep 10:02
· 24 commits to main since this release
v2.1.0
27fc915

AgentDoctor 2.1.0 — Release Notes

Package: @praneeth_54/agentdoctor@2.1.0
Status: Release cut authorized
Release notes: this document

What 2.0.1 was

AgentDoctor 2.0.1 is engineering assurance for AI coding agents:

  • Repository intelligence (graph, impact, architecture signals)
  • Project Brain and knowledge surfaces
  • Safety scan → Safe Fix → verify
  • Change assurance, evidence bundles, and Change Proof integrity
  • Policy / controlled runner (shell=false by default)
  • Workspace isolation
  • Brain + intelligence MCP tools
  • Local dashboard (read-oriented)

It was not a general chatbot and did not claim engineering correctness.

What 2.1.0 adds

An optional Project AI Agent line on top of the same assurance substrate.

AI is opt-in. Core scan / fix / verify / change / Brain / MCP intelligence remain usable without a model provider.

Architecture (invariant)

THE MODEL REASONS.
AGENTDOCTOR PROVIDES PROJECT CONTEXT.
AGENTDOCTOR CONTROLS TOOLS.
AGENTDOCTOR VERIFIES RESULTS.

The model never becomes the source of truth for repository facts. Tool results and repository excerpts are DATA, not SYSTEM instructions.

Surfaces

Surface Role
CLI chat / ask Evidence-backed project Q&A with truth labels
CLI plan / agent Plan + approved coding loop (path-safe tools)
CLI learn Student explain / viva / docs / BUILD_WITH_ME
Dashboard POST /api/chat Ask-only Project Chat; fail-closed if provider=none
MCP agent tools Context, ask, search, path-safe file ops, plan, change verify

Providers

Provider Status
none Default — fail closed for chat surfaces
mock Explicit tests / offline demos only
openai-compatible Optional HTTP (incl. custom base URL)
ollama Optional via OpenAI-compatible endpoint
Native Anthropic / Gemini SDKs NOT IMPLEMENTED

Configure via AGENTDOCTOR_AI_PROVIDER (and related env). Keys are never stored in Brain/evidence.

Highlights by area

Project intelligence

  • Project-aware context packing
  • Project Chat with citations
  • Truth labels: VERIFIED / INFERRED / UNKNOWN / EXTERNAL

Agent tools & coding loop

  • Read / search
  • Create / edit / delete (approval required)
  • Controlled run_command / run_tests
  • PLAN → APPROVAL → EXECUTE → OBSERVE → VERIFY

Safety

  • Path + symlink write protection
  • Workspace isolation (when WorkspaceModel provided) + repo-root binding
  • Mode allowWrites (LEARN cannot write)
  • Approval gates; model cannot self-approve
  • Dangerous command blocking
  • Prompt-injection channel separation
  • Secret redaction
  • Hard limits (tool calls, iterations, wall time, files, context)

Verification

  • Change analysis, evidence, proof, architecture where applicable
  • Optional test run after edits
  • Always: ENGINEERING_CORRECTNESS_NOT_CLAIMED

Student

  • Learn / viva / docs
  • BUILD_WITH_ME → shared coding loop after approval

MCP & dashboard

  • Agent MCP subset (no unrestricted shell)
  • Dashboard ask-only chat; fail-closed on unconfigured AI

Limitations (must ship in release notes)

  • Not an OS sandbox or EDR (EXTERNAL LIMITATION).
  • Native Anthropic / Gemini SDKs NOT IMPLEMENTED.
  • MCP approved=true is trusted-caller input, not cryptographic human identity.
  • MCP does not carry AgentMode (LEARN ban is CLI/loop-scoped).
  • Rich interactive student UI is PARTIAL.
  • Correctness is never guaranteed — integrity ≠ correctness.
  • Do not claim: fully autonomous, 100% correct, completely safe, production guaranteed, or understands everything.

Upgrade notes (after version cut + publish)

npm install -g @praneeth_54/agentdoctor@2.1.0
agentdoctor --version   # expect 2.1.0 after cut

# Optional AI
export AGENTDOCTOR_AI_PROVIDER=mock   # or openai-compatible / ollama
agentdoctor ask "How does authentication work?" .

# Writes still require explicit approval
agentdoctor agent --goal "Add feature" --approve --apply --apply-ops '[...]' .

2.0.1 assurance CLIs (scan, fix, verify, change, evidence, proof, Brain MCP) remain available.

Related docs

Assets 2
Loading