AAHP Orchestrator

June 28, 2026 ยท View on GitHub

VS Code extension - orchestrate GitHub Copilot and Claude Code with AAHP v3 context. Zero questions. Full context. Both agents know what to do.

CI AAHP Verify Security VS Code AAHP v3 License: Apache-2.0

The AAHP Toolchain

Install both packages for the full experience.

PackageWhat it doesWhen to use
aahp-orchestrator - you are hereVS Code extension. Injects AAHP context into Copilot/Claude Code while you code. Live status bar, @aahp chat, sidebar dashboard.Every coding session - your human-in-the-loop assistant.
aahp-runnerAutonomous CLI. Spawns Claude / Gemini / Codex / Copilot agents that implement tasks, run tests, and commit - no human input needed. Schedulable.Overnight / CI - your unattended worker.

Together they cover the full AAHP loop: you plan and guide during the day - the runner works through tasks at night - you wake up to committed progress.


The Problem

Both GitHub Copilot and Claude Code start every session cold. They ask:

  • "What does this project do?"
  • "Which file should I change?"
  • "What conventions do you follow?"
  • "Should I proceed?"

Your AAHP v3 handoff files already have all the answers. This extension makes both agents read them - automatically.


How it works

When you open a workspace with .ai/handoff/MANIFEST.json:

  1. The status bar shows your current phase and active task
  2. A context banner offers to copy AAHP context to clipboard with one click
  3. Use @aahp in any VS Code chat (Copilot or Claude) - full AAHP context is injected automatically as a system prompt before every message
  4. The sidebar dashboard shows all tasks, phase, last agent, and quick-action buttons
  5. Double-click any task in the dashboard to spawn an agent that works on it
  6. Run All Agents to process every repo's top task in parallel

Features

FeatureDetails
@aahp chat participantContext-aware proxy in VS Code chat. Reads MANIFEST.json, injects phase + tasks + conventions + trust as system prompt. No cold starts.
Live status bar[impl] T-003: Fix CORS - updates on every manifest change
Sidebar dashboardFull AAHP state: tasks, phase, agent, quick_context, action buttons
All Open Tasks treeAggregated view of all tasks across all repos, grouped by priority
Auto context bannerOn first editor open: offers to copy AAHP context to clipboard
Multi-repo supportSet aahp.developmentRoot: true to scan all subdirectories for AAHP repos
Agent spawningRun agents for all repos or a single focused repo from the dashboard
Fix Task (play button)Click the play button next to any open task to spawn a targeted agent
Double-click to launchDouble-click any task in Next Steps or task table to start an agent
Dependency checkingWhen launching a task, warns about unresolved dependencies before proceeding
GitHub linksGH buttons next to tasks link directly to GitHub Issues search for that task ID
Agent retryFailed tasks can be retried from the dashboard
Task creationCreate new tasks with title, priority, and dependencies from the dashboard
Task status editingChange task status via dropdown in the task table
Token budget displayShows token usage split between Claude and Copilot backends
Session monitoringDisplays active agent sessions, queued tasks, and completion status

Keyboard Shortcuts

ShortcutCommandDescription
Ctrl+Alt+ACopy Context to ClipboardCopy full AAHP context for pasting into any chat
Ctrl+Alt+DOpen DashboardOpen the AAHP sidebar dashboard
Ctrl+Alt+RRun All AgentsSpawn agents for all repos with ready tasks
Ctrl+Alt+SRun Single RepoSpawn an agent for the currently focused repo

All Commands (Ctrl+Shift+P)

CommandDescription
AAHP: Open DashboardOpen the AAHP sidebar dashboard
AAHP: Copy Context to ClipboardCopy full AAHP context (paste into any chat)
AAHP: Update Manifest ChecksumsRefresh all SHA-256 checksums + line counts in MANIFEST.json
AAHP: Commit SessionStage .ai/handoff/ and commit with one command
AAHP: Set PhaseQuick-pick to switch: research / architecture / implementation / review / fix / release
AAHP: Run All AgentsSpawn agents for every repo with ready tasks
AAHP: Run Agent for Current RepoSpawn an agent for the focused repo only
AAHP: Focus Repo in DashboardSwitch the dashboard focus to a specific repo
AAHP: Set Task StatusChange a task's status (ready/in_progress/done/blocked/pending)
AAHP: Create TaskAdd a new task with title, priority, and dependencies
AAHP: Fix TaskSpawn a targeted agent for a specific task
AAHP: Retry AgentRetry a failed agent with exponential backoff
AAHP: Refresh DashboardRe-scan all repos and reload NEXT_ACTIONS.md

@aahp Chat Commands

Open Copilot Chat or any VS Code chat panel and type @aahp followed by a slash command:

CommandDescription
@aahp /helpShow all available @aahp commands
@aahp /statusShow current status (phase, agent, context)
@aahp /tasksList all tasks with status and priority
@aahp /nextWhat to work on next (top ready task)
@aahp /contextFull injected system prompt for debugging
@aahp /phaseShow or set the current development phase
@aahp /done T-003Mark a task as done

You can also use freeform messages:

@aahp implement the top task
@aahp what's blocking T-003?
@aahp update MANIFEST.json with my progress

The agent already knows your project, phase, conventions, and trust state - no explanation needed.


Dashboard Interactions

The sidebar dashboard supports several interaction patterns:

Play/Run/Retry actions delegate task execution to aahp-runner by launching terminal commands with explicit --repo-path and --task-id arguments.

ActionWhereWhat happens
Click a repo cardRepo Grid sectionFocuses that repo, showing its task table and project details
Click play buttonNext to any open taskSpawns an agent to work on that specific task
Click GH buttonNext to any task with IDOpens GitHub Issues search for that task ID
Double-click a taskNext Steps or task tableSpawns an agent for that task (same as play button)
Change status dropdownTask table, status columnUpdates the task status in MANIFEST.json
Click + New TaskBottom of task tableOpens dialog to create a new task
Click RetryFailed agent cardRe-runs that exact task via aahp-runner
Click Run All AgentsTop of dashboardRuns aahp-runner for all repos
Click Run [repo]Top of dashboard (when focused)Runs aahp-runner for the focused repo/task
Click RefreshNext Steps section headerRe-scans all repos and reloads NEXT_ACTIONS.md
Click section headersAny collapsible sectionCollapse/expand that section

Dependency Warning

When you launch a task that has unresolved dependencies (depends_on in MANIFEST.json), the extension shows a warning dialog listing which dependencies are not yet done. You can choose to proceed anyway or cancel.


Settings

SettingTypeDefaultDescription
aahp.developmentRootbooleanfalseEnable multi-repo mode: scans all subdirectories for .ai/handoff/MANIFEST.json
aahp.rootFolderPathstring""Override the root path to scan for repos (default: workspace root)
aahp.suppressRootPromptbooleanfalseSuppress the initial prompt asking about development root mode
aahp.agentBackendenum"auto"Agent backend for spawning agents. Options: auto (claude > gemini > codex > copilot > sdk), claude, gemini, codex, copilot, sdk
aahp.agentConcurrencyLimitnumber0Max parallel agents when running all repos (0 = unlimited)
aahp.agentMaxRetriesnumber1Max retry attempts on agent failure (uses exponential backoff)

The AAHP sidebar contains two views:

Dashboard (webview)

The main interactive dashboard showing:

  • Agent Control - Run All / Run Single buttons, active agent cards with status
  • Repo Grid - All detected repos with health dots, phase badges, task counts
  • Next Steps - Actionable items parsed from NEXT_ACTIONS.md across all repos
  • Focused Project - Detailed view of the selected repo: task table, status, context
  • Quick Actions - Checksums, Commit, Phase, Context buttons

All Open Tasks (tree view)

A dedicated tree view showing all non-done tasks across all repos, grouped by priority (high/medium/low). Click any task to focus its repo in the dashboard.


Installation

From .vsix

Download the latest .vsix from GitHub Releases, then:

code --install-extension aahp-orchestrator-0.3.5.vsix

Requirements

  • VS Code >= 1.116
  • A workspace with .ai/handoff/MANIFEST.json - create one with aahp init from aahp-runner
  • At least one agent CLI for spawning agents:
    • npm install -g @anthropic-ai/claude-code (Claude Code CLI)
    • npm install -g @google/gemini-cli (Gemini CLI)
    • npm install -g @openai/codex (Codex CLI)
    • gh auth login with a Copilot subscription
  • GitHub Copilot or Claude Code extension for the @aahp chat participant

AAHP v3

This extension reads the following files from .ai/handoff/:

FileUsed for
MANIFEST.jsonPhase, tasks, quick_context, agent, checksums
STATUS.mdCurrent system state
NEXT_ACTIONS.mdPrioritized task list
CONVENTIONS.mdCode style, commit format, tooling
TRUST.mdWhat's verified vs assumed
WORKFLOW.mdAgent pipeline definition

License

Apache-2.0 - elvatis