AGENTIX · OWNED PRODUCTDEVELOPER TOOLING · 2026
§ Owned product

Stack Signal

Customer-hosted AI coding practice analytics

Stack Signal is a local, read-only VS Code extension that turns your AI coding-assistant session logs into actionable insight, across every harness, in one dashboard. Core analysis runs locally; networked features send data only when you enable them.

Apply for the design-partner pilot →All owned products →
0anti-pattern rules
0harnesses · Claude, Copilot, Codex, OpenCode, Xcode
0core telemetry · analysis is local
Honest status: this is early

Stack Signal is in production internally · customer-hosted analytics. It is not yet a finished, generally-available product. What's below describes what it does today, not a roadmap dressed up as a result.

§ Who it is for

A defined user.
A specific problem.

Audience

Engineering leads and developers who use several AI coding assistants and need a private, evidence-based view of how those sessions are working.

Pain

Session logs are split across harnesses, recurring prompt problems stay hidden, and sending internal coding activity to another analytics service is often unacceptable.

§ What it is

Customer-hosted AI coding practice analytics.

Stack Signal reads the AI coding sessions you already generate (Claude, Copilot, Codex, OpenCode, Xcode) from your local logs and turns them into a private dashboard: an activity timeline, how much code came from AI versus your own hands, 45 anti-pattern rules across prompt quality and context management, a Skill Finder that clusters your repeated prompts into reusable skills, context-health audits, and token-spend burndown (the LimitLens view). Core analysis is local and read-only; team roll-up, LimitLens, and optional LLM explanations make network calls when enabled. Stack Signal is tuned to run on Claude Code.

§ The change

Before → system → after.

01 · Before

Separate local logs and no consistent view of prompt quality, context health or AI-assisted output.

02 · Stack Signal

Stack Signal reads five supported harnesses locally and checks sessions against 45 anti-pattern rules.

03 · After

One private dashboard shows activity, repeated skills, context issues and token burndown without uploading core analysis data.

§ How it works

A visible path
through the system.

The steps below describe the current product behaviour from the documented source of truth.

  1. 01

    Read local session logs

    The VS Code extension reads existing Claude, Copilot, Codex, OpenCode and Xcode session logs in read-only mode.

  2. 02

    Normalise the activity

    Sessions from the five harnesses are brought into one local timeline and output view.

  3. 03

    Check 45 rules

    Prompt quality, session hygiene, code review, tool use and context bloat are checked against the built rule set.

  4. 04

    Surface reusable signals

    Skill Finder clusters repeated prompts, while context-health and LimitLens views expose recurring patterns and token use.

§ What it does

Built, not planned.

Any harness, one dashboard

Reads local session logs from Claude, Copilot, Codex, OpenCode and Xcode and unifies them into a single view of how you actually work with AI.

45 anti-pattern rules

Detects prompt smells, session-hygiene issues, weak code review, tool under-use and context bloat, grouped into actionable categories.

Skill Finder

Clusters the prompts you repeat and surfaces what you keep asking AI for, so recurring asks become reusable skills.

Output and spend

Measures AI-generated versus hand-written code volume, and daily token burndown, the LimitLens cost and usage view.

Local and read-only

Core analysis is local; networked features (team roll-up, LimitLens, LLM explanations) are explicitly documented and opt-in.

§ Stack

Deterministic core.
Provider-agnostic edge.

TypeScriptVS Code Extension
§ Practical fit

Deployment, with
the boundaries visible.

Deployment

Install it as a VS Code extension. Core analysis is local and read-only, with no telemetry. Team roll-up, LimitLens and optional LLM explanations make network calls only when you enable them.

Limitations

  • The current log readers cover Claude, Copilot, Codex, OpenCode and Xcode; other harnesses are not claimed.
  • Team roll-up, LimitLens and optional LLM explanations are networked features, so they are not part of the zero-network-call core.
  • Product demonstrations use synthetic sample data.
  • Stack Signal is currently offered as a no-fee design-partner pilot.
  • Upstream is MIT licensed; custom integrations require own deployment.
Next step

See whether Stack Signal
fits your system.

We'll show what exists today, state the limits plainly and decide whether an early product is the right fit.

Apply for the design-partner pilot →See other owned products →
AGENTIX TECH · owned product · In production internally · Customer-hosted analytics