CodeGraph enhanced Secondary Harness for full AI SDLC repository Awareness - Task Manager coordinate and track their own work

[accessvirus]

CodeGraph Feature enhance with Secondary Harness for full AI SDLC repository Awareness - Task Manager coordinate and track their own work — tracking sessions, wiki, knowledge, SDLC records, CI results, and CodeGraph structural context in a single SQLite database

CodeGraph evolution Secondary Harness https://github.com/accessvirus/repository-evolution , under contruction

repository-evolution — Secondary AI SDLC Awareness Harness

Pi Task Manager is a local CLI + API + Streamlit dashboard that AI agents use to coordinate their own work — tracking sessions, wiki, knowledge, SDLC records, CI results, and CodeGraph structural context in a single SQLite database.

This is not a general-purpose app. It is a secondary harness — a coordination substrate that enhances any AI agent's self-awareness of the repository it is working in.

flowchart LR
    subgraph AGENTS [AI Agents]
        CC[Claude Code]
        CX[Codex]
        PI_EXT[Pi Extension\n22 tools]
        UI[Streamlit\n18 pages]
    end

    subgraph SURFACES [Entry Points]
        CLI[taskmgr CLI\nargparse]
        API[FastAPI\n66 routes]
    end

    subgraph INTELLIGENCE [Code Intelligence]
        CG_MCP[CodeGraph MCP\nserver]
        CG_IDX[AST Index\n122 files · 1728 nodes]
    end

    subgraph PERSISTENCE [Persistence]
        DB[(SQLite WAL\n.codegraph/taskmgr.db\n17 tables)]
    end

    CC & CX --> CLI
    CC & CX --> CG_MCP
    PI_EXT & UI --> API
    CLI --> API
    API --> DB
    CLI --> DB
    CG_MCP --> CG_IDX

Loading

---

What It Is

quadrantChart
    title System Role vs Scope
    x-axis Narrow Scope --> Broad Scope
    y-axis Local Tool --> Production System
    quadrant-1 Enterprise Platform
    quadrant-2 Dev-only Broad
    quadrant-3 Simple Local Script
    quadrant-4 Structured Local Harness
    Pi Task Manager: [0.4, 0.2]

Loading

Role	Description
Secondary harness	Augments primary AI agents with persistent SDLC state
Self-using system	Every enhancement uses taskmgr harness start/finish, taskmgr task, taskmgr memory
Single source of truth	One SQLite file at .codegraph/taskmgr.db — all state, no external services
CodeGraph-native	AST-backed structural code intelligence via codegraph_* MCP tools and CLI wrappers

Permanently out of scope: API authentication, JWT/OAuth, webhook HMAC, subprocess sandboxing, rate limiting, multi-tenant isolation, Slack/GitHub Actions/Jira integrations.

---

Architecture

Five Synthesis Layers

flowchart TB
    subgraph L1 [Layer 1 — Structural Awareness]
        DB[database.py\nSQLite WAL + 17 tables]
        RA[repo_awareness.py\nmachine-readable context]
        CGM[codegraph_manager.py\nAST CLI wrappers + evidence]
    end

    subgraph L2 [Layer 2 — Session Coordination]
        HAR[harness.py\nAI session lifecycle]
        TM[task_manager.py\ntask CRUD + ThreadPoolExecutor]
        LM[log_manager.py\noperational log + retention]
    end

    subgraph L3 [Layer 3 — Knowledge Persistence]
        WM[wiki_manager.py\nslugged wiki pages + links]
        KM[knowledge_manager.py\nfacts + decisions + risks]
        MEM[memory_store.py\nkey-value + TTL namespaces]
        TOK[token_tracker.py\nLLM token usage per model]
    end

    subgraph L4 [Layer 4 — SDLC Lifecycle]
        SDLC[sdlc_manager.py\nfeature lifecycle tracking]
        CI[ci_manager.py\nCI profile runner + alerts]
        CTX[context_manager.py\nsession handoff pack — 7 deps]
    end

    subgraph L5 [Layer 5 — Registry Infrastructure]
        AGT[agent_manager.py\nregistry only]
        ALT[alert_system.py\nwrite-only ledger]
        CRON[cron_scheduler.py\nregistry only ⚠️]
        PIPE[pipeline_engine.py\nrun = placeholder ⚠️]
        WH[webhook_handler.py\nno outbound delivery]
        SKL[skill_manager.py\nfilesystem scan]
    end

    L1 --> L2 --> L3 --> L4 --> L5

Loading

[colbymchenry]

Thanks for sharing what you're building!

CodeGraph is intentionally scoped to one thing: a tree-sitter-parsed code-intelligence graph (symbols, edges, files) exposed to AI agents over MCP. Session/task coordination, SDLC tracking, wikis, CI records, and the broader "agent self-awareness harness" you're describing are deliberately out of scope for this project — they sit a layer above CodeGraph, which is exactly where repository-evolution lives.

Since this is a description of your own separate project rather than an actionable change request to CodeGraph itself, I'm going to close this out. Best of luck with the build — glad to have CodeGraph serve as the structural-intelligence layer underneath it.