Skip to content

Changelog

Jump to bottom
McAmner edited this page Jun 2, 2026 · 1 revision

Changelog

All notable changes to mq-agent are documented here.

Format follows Keep a Changelog. Versioning follows Semantic Versioning.

[Unreleased]

Added

  • mq-image-analyze default MCP endpoint (:8766) for visual perception tools.
  • Read-only safety classification for observe_architecture, image_ocr, analyze_*, extract_*, reverse_* and compare_* MCP tool families.
  • Tests and docs for routing mq-agent run-tool observe_architecture and mq-agent run-tool image_ocr to mq-image-analyze.
  • mq-agent review file|diff|repo --architecture-image <path> / --visual <path> to pass visual_architecture_observation.v1 context from mq-image-analyze into mq-mcp architecture reviews.

[v1.3.0] — 2026-05-31

Added

  • Architecture-memory context in review workflows — list_architecture_decisions called automatically after review findings; shown as compact panel when tool is available; silent when not present in mq-mcp
  • --fast flag on review file/diff/repo — passes fast=True to mq-mcp; mq-mcp routes to Class A tools internally (model-selection policy)
  • mq-agent learn status — check mq-mcp learn system availability
  • mq-agent learn search <query> — search mq-mcp learned review patterns (read-only)
  • mq-agent learn explain <pattern-id> — fetch pattern explanation (read-only)
  • MultiMCPBridge._call_optional_tool() — like _call_required_tool but returns None when tool is absent, for use with optional enrichment
  • MultiMCPBridge.get_architecture_decision(id) — fetch single decision by ID
  • 24 new tests in tests/test_v1_3_features.py (292 total)

Changed

  • _review_flags() includes fast key; existing flag dicts updated in tests
  • _run_review() accepts optional bridge for arch-context rendering
  • COMMAND_SURFACE.md — review section extended with --fast; learn section added
  • EXAMPLES.md — --fast and arch-context examples; learned patterns section added
  • ROADMAP.md — v1.1.0 remaining items ticked; v1.3.0 section added; release map updated

[v1.2.0] — 2026-05-31

Added

  • mq-agent memory search <query> — searches mq-mcp semantic memory via search_semantic_memory; renders key/excerpt table; degrades with a clear error when mq-mcp v1.4.0 tool is absent; --json supported
  • mq-agent memory store <key> <value> --approve — writes to mq-mcp semantic memory via store_semantic_memory (Class C write tool); requires --approve; --dry-run shows planned call without executing; --json supported
  • MultiMCPBridge.search_semantic_memory() and store_semantic_memory() bridge methods
  • mq-agent mcp status — extended to show semantic memory item count and contract freshness (validate_orchestration_contract) when those tools are available in mq-mcp
  • 16 new tests in tests/test_semantic_memory_cli.py and tests/test_mcp.py (268 total)

Changed

  • COMMAND_SURFACE.md — memory section updated with search and store entries; mcp status notes updated with enrichment
  • EXAMPLES.md — semantic memory section extended with search/store examples
  • ROADMAP.md — v1.2.0 items ticked; v1.1.0 marked Done in release map

[v1.1.0] — 2026-05-31

Added

  • mq-agent review file/diff/repo — pass-through review orchestration via mq-mcp (review_file, review_diff, review_repo); severity findings rendered without local reinterpretation
  • --security, --architecture, --risk flags forwarded to mq-mcp review contracts
  • --dry-run on all review commands — shows what would be called without executing
  • validate_orchestration_contract check in mq-agent doctor — verifies mq-mcp orchestration contract is available
  • 3 new dry-run tests in tests/test_review_cli.py (252 total)

Changed

  • COMMAND_SURFACE.md — review section updated with --dry-run entry
  • EXAMPLES.md — review section extended with dry-run examples
  • ROADMAP.md — v1.1.0 items ticked; arch-memory and model-selection deferred to v1.2.0 (require mq-mcp v1.4.0+); release map updated to In Progress

[v1.0.0] — 2026-05-26

Added

  • mq_agent/cli/render.py — Rich rendering helpers (print_steps, print_tool_spec, print_mcp_tool_table, print_swarm_result) extracted from main.py; each takes an explicit console arg for testability
  • mq_agent/core/diagnostics.py — run_checks() and required_checks_pass() for doctor command; no CLI/Rich dependencies
  • tests/test_diagnostics.py — 8 tests for environment health checks

Changed

  • Planner now reads model via load_config().effective_model() instead of os.environ directly; respects ~/.mq-agent/config.json settings
  • main.py reduced from 1381 → 1270 lines; all rendering and doctor logic delegated
  • ARCHITECTURE.md rewritten: full module map, two parallel runtime modes, config schema, safety model table, module boundaries
  • EXAMPLES.md extended to cover all commands: task, browser, swarm, memory, mcp, score, signal
  • pyproject.toml description updated to "workflow orchestration runtime"

v1.0.0 platform summary

237 tests · green CI · protected main · full docs · complete examples · stable contracts locked in v0.9.0 · versioned releases v0.1.0–v1.0.0

[v0.9.0] — 2026-05-26

Added

  • mq_agent/config.py — MqAgentConfig dataclass with load_config() and save_config(). Schema: safety_mode, model, dry_run, working_dir. Reads from ~/.mq-agent/config.json; falls back to built-in defaults. effective_model() prefers MQ_AGENT_MODEL env var.
  • 7 new contract tests in tests/test_orchestration_contract.py:
    • test_plan_step_fields_are_stable — locks PlanStep field set
    • test_step_result_fields_are_stable — locks StepResult field set
    • test_agent_manifest_fields_are_stable — locks AgentManifest field set
    • test_agent_state_to_dict_shape_is_stable — locks to_dict() keys
    • test_agent_manifest_allowed_tools_must_be_subset_of_registry
    • test_run_task_tool_delegates_to_task_runner
    • test_tool_registry_has_no_positional_only_params

Changed

  • PlanStep (core/state.py) — added docstring clarifying it is the Executor loop step model, distinct from StepResult in task_runner.
  • StepResult (core/task_runner.py) — added docstring clarifying it is the declarative YAML task step model, distinct from PlanStep.
  • SwarmRunner (core/swarm.py) — expanded docstring: documents it as a separate runtime mode from the Executor loop, not a competing impl.

Test count

218 tests (was 211).

[v0.8.0] — 2026-05-24

Added

  • mq-agent swarm list — list available swarm configs with agents and safety classes
  • mq-agent swarm plan <config> — show participating agents; no API calls needed
  • mq-agent swarm run <config> [path] — execute a named swarm, unified report
  • mq-agent swarm audit [path] — shorthand: audit + signal + docs swarm
  • mq-agent swarm release-check [path] — shorthand: CI + audit + release swarm
  • mq_agent/core/swarm.py — AgentManifest, SwarmConfig, AgentResult, SwarmResult, SwarmRunner
  • mq_agent/agents/swarm_registry.py — built-in swarm configs: audit, release-check, ci
  • Each AgentManifest declares: purpose, safety_class, allowed_tools, requires_approve, output_contract, failure_behavior
  • Dry-run mode requires no API key and skips all agent execution
  • --approve gate for write-capable agents (release manifest)
  • tasks/swarm_audit.yaml — declarative swarm audit task workflow
  • tests/test_swarm.py — 29 tests covering manifests, configs, runner, registry, CLI
  • 190 tests total

[v0.7.0] — 2026-05-24

Added

  • mq-agent browser inspect <url> — fetch and parse URL metadata: title, description, headings, links, word count
  • mq-agent browser summarize <url> — plain-text content summary from a URL
  • mq-agent browser verify-release <url> — verify release page fields; --tag <v> checks expected version tag
  • mq_agent/tools/browser_tools.py — fetch_url, inspect_url, summarize_url, verify_release_url
  • _HTMLExtractor — stdlib-only HTML parser (title, description, h1/h2, links, text)
  • URL safety gate: blocks file://, ftp://, data:, javascript: schemes
  • Browser tools registered in TOOL_REGISTRY — usable in task YAML workflows
  • tasks/browser_verify.yaml — declarative task for release page verification
  • tests/test_browser.py — 27 tests covering URL safety, HTML parsing, inspect/summarize/verify, and CLI commands
  • 161 tests total

[v0.6.0] — 2026-05-24

Added

  • mq-agent task list — list available YAML task workflow files
  • mq-agent task list --json — machine-readable task list with name, description, step count
  • mq-agent task run <name> — execute a declarative YAML workflow via the tool registry
  • mq-agent task run <name> --dry-run — preview steps without execution (default: False)
  • mq-agent task run <name> --json — machine-readable step results
  • mq_agent/core/task_runner.py — TaskStep, StepResult, Task, load_task, run_task, find_task_files
  • Task lookup by filename stem OR internal YAML name: field — both audit and repo-audit resolve correctly
  • tests/test_task_runner.py — 18 tests covering load_task, run_task, find_task_files and CLI commands

[v0.5.2] — 2026-05-24

Added

  • mq-agent mcp start — start mq-mcp server in background (SSE transport, PID file managed)
  • mq-agent mcp stop — stop the background mq-mcp server
  • mq_agent/mcp/manager.py — PID-file-based process lifecycle: start(), stop(), is_running(), read_pid()
  • mq-agent mcp status now shows process state (running/not running) from PID file alongside HTTP reachability
  • mq-agent mcp status --json includes process_running and pid fields
  • mq-agent mcp start/stop --json — machine-readable output
  • 14 new tests in tests/test_mcp.py covering manager module and CLI commands

Changed

  • mq-agent mcp status — not-running state now shows "Start with: mq-agent mcp start"
  • MCPBridge.not_reachable_message() — now suggests mq-agent mcp start instead of raw uv command

[v0.5.1] — 2026-05-24

Added

  • mq-agent memory doctor — diagnose environment with per-item status and actionable fixes
  • mq-agent memory doctor --json — machine-readable diagnostics
  • mq-agent memory status --json — machine-readable status output
  • DoctorReport and DiagnosticItem dataclasses in mq_agent/memory/semantic.py
  • doctor() function in mq_agent/memory/semantic.py
  • tests/test_memory_cli.py — 18 CLI-level tests covering dry-run default, approval gate, JSON output and doctor diagnostics
  • Example output for memory doctor in docs/SEMANTIC_MEMORY.md and README

Changed

  • README semantic memory section expanded with verified output and memory doctor example
  • docs/COMMAND_SURFACE.md updated with memory doctor and --json variants

[v0.5.0] — 2026-05-24

Added

  • mq-agent memory status — check vector store ID and repo-signal availability
  • mq-agent memory build . — dry-run semantic upload preview (safe default)
  • mq-agent memory refresh . --approve — upload semantic memory (gate required)
  • mq_agent/memory/semantic.py — SemanticMemoryStatus, get_vector_store_id(), status(), build()
  • docs/SEMANTIC_MEMORY.md — commands, environment, safety model and failure states
  • scripts/smoke-memory.sh — smoke test for memory status and build dry-run

Safety

  • memory build defaults to --dry-run; no upload without --no-dry-run
  • memory refresh requires --approve; exits 1 without it
  • Missing vector store and missing repo-signal reported explicitly, never silent

Changed

  • README updated to v0.5.0 with semantic memory section
  • Roadmap updated: v0.5.0 semantic memory marked complete; autonomous loop/browser moved to Later

[v0.4.1] — 2026-05-23

Added

  • docs/COMMAND_SURFACE.md — canonical command-count reference for mq-agent, MCP commands, mqlaunch menu items, direct prompt commands and smoke-test coverage
  • Docs consistency checks guard mqlaunch command counts and require all six direct prompt commands to be documented
  • mq-agent docs-audit . — CLI command exposing DocsAgent (implementation existed but had no registered entrypoint)
  • MQ_AGENT_MODEL env var — overrides the planner model (default: gpt-4o)
  • MQ_AGENT_VERIFIER_MODEL env var — overrides the verifier model (default: gpt-4o-mini)
  • memory field on AgentState — Memory wired into the agent lifecycle; AuditAgent loads prior context and saves summary after each run

Removed

  • mq_agent/cli.py — dead parallel click-based CLI (never registered as entrypoint, click was an undeclared dependency)
  • mq_agent/memory/ — empty directory left over from a refactor

Fixed

  • Safety gate _is_destructive() matched keywords in step descriptions ("format output", "remove duplicates") — now checks run_command args only
  • find_files() parameter renamed suffix → pattern to match AI planner output; default updated from ".py" to "*.py"
  • TUI version label hardcoded to v0.1.0 — now reads __version__ from package metadata
  • TUI command list was stale (7 entries, pre-v0.3.0) — updated to all 12 current commands
  • TUI subprocess.run() blocked the Textual event loop — replaced with asyncio.create_subprocess_exec with streaming output
  • __version__ now reads from package metadata instead of a hardcoded string
  • Smoke test removed release-check --dry-run (requires OPENAI_API_KEY); replaced with repo-summary .

Changed

  • README and GitHub Pages updated to v0.4.1
  • docs/index.html restructured: 6-card architecture grid, commands split by API-key requirement, links as 2-column grid
  • docs/COMMAND_SURFACE.md updated with docs-audit and corrected smoke-test coverage table

[v0.4.0] — 2026-05-23

Added — mqlaunch integration

  • mq-agent-menu.sh (macos-scripts) — 12-item agent menu: REPO ANALYSIS (1-4), AI COMMANDS (5-8), ENVIRONMENT (9-10), MCP LOCAL TOOLS (11-12)
  • mqlaunch agent ... command surface for doctor, score, audit, release-check, MCP status and MCP tools
  • 6 direct prompt commands wired into mqlaunch: agent score, agent audit, agent doctor, agent release-check, agent mcp-status, agent mcp-tools
  • scripts/smoke-mqlaunch.sh — smoke test that verifies mqlaunch agent ... can reach mq-agent without a live session
  • docs/MQLAUNCH_INTEGRATION.md — bridge architecture, available commands, and smoke-test usage

Changed

  • README: v0.4.0 status section; mqlaunch bridge section added; Roadmap updated
  • ROADMAP: v0.4.0 marked done; v0.5.0 is autonomous loop mode

Version

  • Bumped to 0.4.0

[v0.3.0] — 2026-05-23

Added — local tool orchestration

  • mq_agent/tools/mcp_registry.py — MCPToolSpec dataclass, classify_tool_name(), and five safety classes: read-only, write-capable, subprocess, dangerous, unknown
  • mq-agent mcp status — checks mq-mcp reachability and reports tool counts by safety class
  • mq-agent mcp tools — lists all MCP-discovered tools with safety class and description
  • mq-agent tools --describe <name> — shows description, schema, examples and safety class for a tool
  • mq-agent tools --mcp — includes discovered MCP tools alongside built-in tools
  • mq-agent tools --json — JSON output for built-in and/or MCP tool listing
  • mq-agent run-tool <tool> — runs a specific MCP tool through mq-agent safety gates; --arg key=value for arguments
  • docs/MCP_INTEGRATION.md — architecture and usage guide for mq-mcp integration
  • docs/TOOL_ROUTING.md — safety classification reference and routing rules

Changed

  • mq_agent/tools/mcp_bridge.py — added list_tool_specs(), describe_tool(), not_reachable_message(), and _fetch_tools_raw() for structured metadata; backward-compatible
  • mq-agent tools — now accepts --describe, --mcp, and --json flags
  • README updated with v0.3.0 status, new CLI commands, and test count (70, was 37)
  • ROADMAP updated: v0.3.0 marked complete, v0.4.0 is mqlaunch integration

Safety

  • Unknown MCP tools are blocked by default — no classification, no execution
  • Write-capable and subprocess tools require --approve to run
  • Dangerous tools require --dangerous to run
  • Dry-run mode previews the intended MCP call without contacting mq-mcp
  • Fail-closed: when mq-mcp is unavailable, run-tool exits with a clear error and startup instructions

Tests

  • tests/test_mcp.py — 33 new tests covering: name classification (17), MCPToolSpec construction and round-trip (9), bridge behavior when unavailable (6), safety count validation (1)
  • Total: 70 tests pass (was 37) — no OpenAI API key or mq-mcp server required

Version

  • Bumped to 0.3.0

[v0.2.4] — 2026-05-23

Added — docs sync + self-checking

  • scripts/check-docs-consistency.sh — verifies version sync across VERSION, pyproject.toml, README badge, README status section, CHANGELOG, and docs/index.html; blocks stale score strings (70/100, 8/16) and old version references
  • release-check.sh — pre-release gate: tests, lint, docs consistency
  • Docs consistency step added to tests.yml CI workflow
  • ROADMAP rewritten to reflect actual done/next/later state (was showing everything as unchecked)

Fixed — v0.2.4

  • docs/index.html version badge corrected from v0.2.1 to v0.2.4
  • README status badge and section heading updated from v0.2.3 to v0.2.4
  • scripts/check-install.sh: doctor now uses || true (matches CI, avoids failure without OPENAI_API_KEY)
  • Version bumped to 0.2.4 in VERSION and pyproject.toml

[v0.2.3] — 2026-05-23

Added — credibility polish

  • Proof section in README — lists what is actually verified
  • scripts/check-install.sh — local install smoke test
  • .github/workflows/install-smoke.yml — CI smoke test (no API key)
  • README Demo section updated with real 100/100 output
  • README status section updated to v0.2.3

Changed — v0.2.3

  • Version bumped to 0.2.3

[v0.2.2] — 2026-05-23

Added — demo polish

  • docs/DEMO.md — end-to-end walkthrough with real command output
  • docs/COMMANDS.md — full command reference with safety modes and flags
  • docs/SAFETY_CONTRACT.md — what the agent may/must not do, blocked patterns, audit trail
  • docs/MQ_ECOSYSTEM.md — mqlaunch, mq-agent, mq-hal, mq-mcp, repo-signal integration map
  • SKILLS.md + skills/ — four skill definitions: repo-audit, release-readiness, signal-assessment, ci-diagnosis
  • README: 30-second demo section, Use cases, tiered install (dev/GitHub/PyPI)
  • GitHub: 11 topics set

Fixed — v0.2.2

  • read_file() accepts file_path kwarg (GPT-4o natural argument compat)
  • signal_tools.py ruff + mypy clean (UP035, no-redef, import sort)
  • mcp_bridge list_tools return type narrowed to list[str]
  • planner/verifier: json.loads(content or "{}") — str | None guard
  • TUI: BINDINGS type aligned with Textual App base class
  • TUI: Log() markup kwarg removed (not supported)
  • CI: --system flag removed from uv install; [signal] extra added
  • python-dotenv auto-load from ~/mq-agent/.env and ./.env

Changed — v0.2.2

  • Version bumped to 0.2.2

[v0.2.0] — 2026-05-23

Added — repo-signal integration

  • mq_agent/tools/signal_tools.py — four tool wrappers around repo-signal's public API:
    • repo_scan — full repo scan: project type, languages, tooling, entry points, focus areas
    • repo_readme_score — README quality score 0–100 with visual bar and per-check breakdown
    • repo_publish_checklist — publish readiness checklist with grouped checks and action hints
    • repo_analyze — full repo-signal analysis report (markdown)
    • repo_signal_json — combined structured dict for use by agents and CI
  • mq_agent/agents/signal_agent.py — SignalAgent: static signal pass + AI improvement plan
  • mq-agent signal [path] — full assessment with overall score, README gaps, focus areas, AI plan
  • mq-agent score [path] — instant README score + publish checklist (no AI, no API key needed)
  • All four signal tools registered in TOOL_REGISTRY and SAFE_TOOLS
  • repo-signal added as [signal] optional extra in pyproject.toml
  • mq-agent doctor now shows repo-signal availability
  • 11 new tests in tests/test_signal_tools.py (37 total)

Changed — v0.2.0

  • Version bumped to 0.2.0

Fixed — signal_tools syntax

  • *expr or fallback syntax error inside list literals in signal_tools.py

[v0.1.0] — 2026-05-23

Added — core orchestration layer

  • Planner — OpenAI gpt-4o backed structured plan generation from goals
  • Executor — tool registry routing with dry-run and safety gate enforcement
  • Verifier — per-step AI verification using gpt-4o-mini
  • Memory — two-level memory: session (ephemeral) and persistent (JSON on disk)
  • SafetyGate — four safety modes: read-only, suggest, execute, dangerous
  • AgentState — structured state object tracking plan, context and progress
  • Git tools: git_status, git_log, git_diff, git_branch, git_remote
  • Shell tools: run_command with blocked dangerous pattern list
  • Repo tools: repo_summary, list_files, read_file, find_files
  • MCP bridge: HTTP bridge to mq-mcp server
  • AuditAgent, ReleaseAgent, CIAgent, DocsAgent
  • Typer CLI with audit, plan, release-plan, release-check, repo-summary, run, fix-ci, doctor, tools, tui
  • All commands support --dry-run and --json
  • Textual TUI with sidebar navigation and live log output
  • GitHub Actions CI: tests + lint + type check on every push
  • 26 unit tests, no OpenAI calls required
  • Declarative task YAML: release.yaml, audit.yaml, fix_ci.yaml

Fixed — v0.1.0

  • git_log parameter renamed n → limit to match GPT-4o natural arg generation
  • git_branch, git_remote, find_files, which added to SAFE_TOOLS
  • Executor: read-only mode now continues past step failures instead of stopping

[Unreleased]

Planned for v0.2.0

  • repo-signal integration
  • repository audit scoring
  • release readiness scoring
  • semantic repository memory

mq-agent

Clone this wiki locally