Telos Guardian

Overview

Guard a project's purpose by ensuring every proposed feature aligns with the project's reason for existing (its telos) and is practically achievable given the current codebase. This skill operates in two modes: Init (capture the telos) and Assess (evaluate feature alignment).

When to use

• The task is to capture a project’s purpose hierarchy in

  TELOS.md

  or assess a proposed feature against it.
• The user is planning new work and needs an alignment check before implementation.
• The request involves “should we build this?”, “does this fit the project?”, or initializing telos for a new project.
• The output should clarify purpose, boundaries, user experience intent, and feature fit.

Do NOT use when:

• The task is a small bug fix, typo, or mechanical refactor with no product-scope change.
• The feature is already clearly in-bounds and the user is asking for implementation help, not alignment review.
• The user explicitly wants to skip telos or product-alignment analysis.

Response format

Always structure the final response with these top-level sections, in this order:

1. Summary — state the task, scope, and main conclusion in 1-3 sentences.
2. Decision / Approach — state the key classification, assumptions, or chosen path.
3. Artifacts — provide the primary deliverable(s) for this skill. Use clear subheadings for multiple files, commands, JSON payloads, queries, or documents.
4. Validation — state checks performed, important risks, caveats, or unresolved questions.
5. Next steps — list concrete follow-up actions, or write

   None

   if nothing remains.

Rules:

• Do not omit a section; write

  None

  when a section does not apply.
• If files are produced, list each file path under Artifacts before its contents.
• If commands, JSON, SQL, YAML, or code are produced, put each artifact in fenced code blocks with the correct language tag when possible.
• Keep section names exactly as written above so output stays predictable across skills.

Workflow

Step 0: Detect mode

Check if a

TELOS.md

file exists at the project root.

• If NO

  TELOS.md

  exists -> go to Mode A: Init
• If

  TELOS.md

  exists -> go to Mode B: Assess

---

Mode A: Init (Capture the Telos)

Guide the user through a structured interview to capture the project's purpose hierarchy across all 4 levels. Ask questions one at a time, not in batches. Prefer multiple-choice when appropriate, but open-ended is fine for vision questions.

A1. L4 Purpose — Why does this project exist?

Ask the user:

1. What is the ultimate purpose of this project? What problem does it solve, and for whom? Push for specificity — reject generic answers like "help users" or "make things easier." Ask "why?" until you reach a concrete, meaningful purpose.
2. Who are the beneficiaries? Primary users, secondary stakeholders, ecosystem participants.
3. What does success look like? Concrete, measurable success metrics. Not "lots of users" but "1000 monthly active users within 6 months" or "reduce deployment time from 2 hours to 10 minutes."
4. What constraints exist? Technical (must run on X), business (budget, timeline), philosophical (open source, privacy-first), regulatory.

A2. L3 Experience — What should it feel like?

Ask the user:

1. What are the key user journeys? Walk through 2-3 critical paths a user takes from start to goal. Be concrete: "User opens the app, sees X, clicks Y, gets Z."
2. What experience principles matter? Speed over polish? Simplicity over power? What trade-offs define the product personality?
3. What feedback loops exist? How will you know users are succeeding? What analytics, feedback mechanisms, or signals matter?

A3. L2 Contract — What are the boundaries?

Ask the user:

1. What are the major interfaces? APIs, component boundaries, service contracts, data models. What talks to what?
2. What external systems integrate? Databases, third-party APIs, auth providers, payment systems.
3. What data contracts exist? Key data shapes, validation rules, invariants that must hold.

If the project is early-stage and these don't exist yet, capture what the user envisions. Mark uncertain items as "TBD" in the output.

A4. L1 Function — What does the code look like?

Ask the user:

1. What are the core modules? The main functional areas of the codebase.
2. What patterns does the codebase follow? Architecture style (MVC, hexagonal, microservices), language conventions, state management approach.
3. What is the testing strategy? Unit tests, integration tests, E2E? What framework? What coverage expectations?

If the project doesn't exist yet, capture the intended patterns.

A5. Write TELOS.md

Produce a

TELOS.md

file at the project root using this format:

# TELOS — [Project Name]

> [One-sentence purpose statement]

## L4: Purpose

**Why this project exists:** [Specific, concrete purpose — not generic]

**Beneficiaries:**

- [Primary]: [who and why]
- [Secondary]: [who and why]

**Success metrics:**

-
-

**Constraints:**

- [Constraint 1]
- [Constraint 2]

## L3: Experience

**Key user journeys:**

### Journey 1: [Name]

[Step-by-step description]

### Journey 2: [Name]

[Step-by-step description]

**Experience principles:**

- [Principle 1]
- [Principle 2]

**Feedback loops:**

- [Loop 1]
- [Loop 2]

## L2: Contract

**Major interfaces:**

-
-

**Integration points:**

- [System 1]: [how it connects]

**Data contracts:**

- [Contract 1]: [shape and invariants]

## L1: Function

**Core modules:**

-
-

**Patterns:**

- Architecture: [style]
- Language: [conventions]
- State management: [approach]

**Testing strategy:**

- [Framework]: [coverage expectations]

Save this file and confirm with the user that it captures their intent. Make adjustments as needed until the user is satisfied.

---

Mode B: Assess (Feature Alignment)

When a feature is proposed and

TELOS.md

exists, assess alignment and feasibility through conversation.

B1. Read the TELOS.md

Load the project's

TELOS.md

. Internalize all 4 levels — you need the full context to evaluate alignment.

B2. Understand the proposed feature

If the user hasn't fully described the feature, ask clarifying questions until you understand:

• What the feature does
• Who it's for
• What problem it solves

B3. Scan the codebase

Examine the project structure to understand:

• Key directories and modules
• Dependencies and frameworks in use
• Existing patterns (architecture style, naming conventions, testing approach)
• Related code that the feature would interact with or extend

This is an architecture scan, not a deep code review. Focus on whether the feature fits naturally into the existing structure or would require significant refactoring.

B4. Assess alignment at each level

Walk through each level conversationally:

L4 Purpose alignment:

• Does this feature serve the stated ultimate purpose?
• Does it support the defined success metrics?
• Does it respect the stated constraints?
• Could it undermine or distract from the core purpose?

L3 Experience alignment:

• Which user journeys does this feature affect?
• Does it align with the stated experience principles?
• Does it introduce new journeys that complement or conflict with existing ones?

L2 Contract alignment:

• How does this feature interact with existing interfaces?
• Does it require new contracts, or extend existing ones?
• Does it respect current system boundaries?

L1 Function feasibility:

• Does the codebase architecture support this feature naturally?
• What modules would need to change?
• Does it fit the existing patterns, or would it require introducing new ones?
• Are the necessary dependencies available?

B5. Deliver your assessment

Provide a clear, conversational assessment covering:

1. Purpose alignment — how well the feature serves the telos
2. Practical feasibility — how naturally it fits the codebase
3. Recommendation — one of:
   • Proceed — strongly aligned and feasible
   • Proceed with caveats — aligned but has considerations to address first
   • Reconsider — misaligned with purpose, or would require rethinking the approach

If recommending "proceed with caveats" or "reconsider," explain specifically what the concerns are and what would need to change.

---

Checklist

• Checked for existing

  TELOS.md

  at project root
• (Init) Asked L4 questions — purpose, beneficiaries, metrics, constraints
• (Init) Asked L3 questions — journeys, experience principles, feedback
• (Init) Asked L2 questions — interfaces, integrations, data contracts
• (Init) Asked L1 questions — modules, patterns, testing
• (Init) Wrote

  TELOS.md

  to project root and confirmed with user
• (Assess) Read existing

  TELOS.md

• (Assess) Understood the proposed feature
• (Assess) Scanned codebase architecture
• (Assess) Evaluated alignment at all 4 levels (L4, L3, L2, L1)
• (Assess) Delivered clear recommendation (proceed / caveats / reconsider)

Example

Mode A: Init — New SaaS project

User: "I want to start building an invoice management tool for freelancers."

Agent interview (abbreviated):

L4 Purpose: "Why does this need to exist? What's broken about current invoicing for freelancers?"

User: "Freelancers waste 3-4 hours/week on invoicing. Existing tools are bloated and expensive. We want something minimal that gets invoices sent in under 60 seconds."

L4 Success metrics: "How will you measure success?"

User: "Time to send first invoice under 60 seconds, 80% of users send at least 1 invoice/week within 30 days."

[Agent continues through L3, L2, L1...]

Output

TELOS.md

(excerpt):

# TELOS — QuickBill

> Eliminate invoicing friction for freelancers so they spend time on billable
> work, not admin.

## L4: Purpose

**Why this project exists:** Freelancers waste 3-4 hours per week on invoicing.
Existing tools (FreshBooks, Wave) are over-featured and slow. QuickBill gets an
invoice sent in under 60 seconds.

**Beneficiaries:**

- Primary: Solo freelancers (designers, developers, writers)
- Secondary: Small agencies (2-5 people) who need simple shared invoicing

**Success metrics:**

- Time to send first invoice: < 60 seconds
- Weekly active invoicing rate: 80% of users within 30 days
- User retention at 90 days: > 60%

**Constraints:**

- Must work offline (freelancers on unreliable connections)
- No monthly fees (one-time purchase or freemium)
- GDPR compliant (European freelancer market)

Mode B: Assess — Feature proposal against existing telos

User: "I want to add a built-in time tracker to QuickBill."

Agent assessment:

L4 Purpose alignment: QuickBill's purpose is to eliminate invoicing friction. A time tracker is adjacent — it could feed tracked hours directly into invoices, reducing friction further. However, the constraint says "must work offline" and real-time tracking adds sync complexity. The constraint also emphasizes minimalism — adding time tracking risks the "over-featured" problem you're trying to solve.

L3 Experience alignment: Your key journey is "open app, create invoice, send in 60 seconds." A time tracker introduces a separate workflow (start timer, track, stop, then invoice). This could complicate the primary journey unless it's deeply integrated.

L2 Contract alignment: Currently your data model has Invoice and LineItem entities. A time tracker would need TimeEntry, Project, and potentially Client entities. This roughly doubles your data model surface area.

L1 Function feasibility: Your codebase is a simple React app with local storage. Adding real-time tracking would need background workers or service workers for offline tracking, which is a significant architecture addition.

Recommendation: Proceed with caveats. The feature could serve the purpose if scoped tightly — "convert tracked time to invoice line items in one click." But if it becomes a full time-tracking app, it risks purpose drift. Suggest: start with a minimal "quick log hours" feature that populates invoice line items, not a real-time timer.

Common mistakes

Mistake	Fix
Generic TELOS.md ("we want to build good software")	Push for specificity. Ask "why?" repeatedly. Reject vague purpose statements.
Skipping the codebase scan during assessment	The feasibility check is essential. Always examine project structure before delivering a recommendation.
Rubber-stamping features as "aligned"	Genuinely challenge features. If a feature doesn't clearly serve the purpose, say so. The skill's value is in the features it catches, not the ones it approves.
Asking all questions at once in Init mode	Ask one question at a time. Let the user think and respond before moving to the next.
Treating this as code review	This is pre-implementation alignment, not post-implementation review. Focus on purpose and architecture fit, not code quality.
Writing TELOS.md without user confirmation	Always show the draft and ask if it captures their intent before saving.

Quick reference

Situation	Action
No TELOS.md exists	Run Mode A: Init interview (L4 -> L3 -> L2 -> L1), write TELOS.md
TELOS.md exists + feature proposed	Run Mode B: Read telos, scan codebase, assess alignment at all 4 levels
Feature clearly aligned	Recommend "proceed" with brief rationale
Feature partially aligned	Recommend "proceed with caveats" and explain what to address
Feature misaligned with purpose	Recommend "reconsider" and explain the conflict
User wants to update the telos	Re-run relevant sections of Mode A, update TELOS.md

Key principles

1. Purpose over features — Every feature must serve the project's reason for existing. A feature that doesn't align with the telos is waste, regardless of how technically interesting it is.

2. Specificity over generality — Vague purpose statements ("help users," "improve productivity") are useless for alignment checks. The skill must push for concrete, falsifiable statements about why the project exists and what success looks like.

3. Honest assessment over approval — The skill's value comes from catching misaligned features before they're built. If a feature doesn't serve the purpose, say so clearly. False agreement wastes the user's time and budget.

4. Practical feasibility matters — Purpose alignment alone isn't enough. A feature can serve the telos beautifully but be impractical given the current codebase. Always scan the architecture before recommending.

5. One question at a time — During Init, don't overwhelm the user. Ask a single focused question, wait for the answer, then proceed. This produces better, more thoughtful answers.