🎯 Goal: Finish Line Defined

Define what "done" looks like and how you'll prove it, before you write a line of code.

Activation

When this skill activates, output:

🎯 Goal: Finish line defined.

Then walk the user through the five fields in order, one prompt at a time or as a single template: whichever fits the flow. Fill in what the request already tells you; ask only for what's missing. Put the most weight on VERIFICATION: it is the field that is required, the field that is skipped most often, and the field that turns "I think it's done" into "I can show it's done." Do not let a goal close without a concrete, checkable verification criterion.

Context Guard

The Five Fields

A goal is a task with a finish line you can stand on. Five fields:

1. TASK: what to do. The concrete action, in one sentence. Required.
2. WHY: why it matters / who it's for. Optional, but it keeps scope honest and kills gold-plating. Skip it if it's obvious.
3. OUTCOME: the exact finished result, the definition of done. Not "improve X" but the specific end state that exists when you're finished. Required.
4. CONSTRAINTS: what must or must not be done. Boundaries: no new dependencies, don't change the public API, stay under budget. Optional; list as many as apply.
5. VERIFICATION: how you'll PROVE the outcome is complete. A concrete, observable, checkable criterion: a command to run, an output to see, a behavior to reproduce. Required, and load-bearing.

A goal without a concrete VERIFICATION criterion is incomplete, not merely unpolished. "It works," "done," and "tests pass" are not verification, they assert success without naming what you'd observe to confirm it. That gap is exactly how a green build ships the wrong behavior. State the check, not the vibe.

How It Works

The skill shells out to the goal engine for parsing and validation (the same engine the Architect stage uses) so a drafted goal is checked the same way everywhere.

# Print the blank five-field template
python -m memstack_skill_loader.goal template

# Validate a drafted goal (reads the labeled text, reports errors + warnings)
python -m memstack_skill_loader.goal validate

validate enforces the rules: TASK, OUTCOME, and VERIFICATION must be present; a present-but-vacuous VERIFICATION (e.g. "it works") comes back as a warning telling you to state a concrete, checkable criterion. Errors mean the goal isn't ready to execute; warnings mean tighten it before you trust it.

Note: the engine module (goal_engine.py) exists today; the thin CLI entry (python -m memstack_skill_loader.goal) is wired in a later registration step. If the command isn't callable yet, don't block, guide the user through the five fields by hand and show the template inline (below). The discipline is the point; the CLI is a convenience.

Inline template (fallback when the CLI isn't available):

TASK: (what to do)
WHY: (why it matters / who it's for - optional)
OUTCOME: (the exact finished result - what 'done' looks like)
CONSTRAINTS: (what must or must not be done - optional, one per line)
VERIFICATION: (how you'll prove the outcome is complete - a concrete, checkable criterion)

Quick Commands

• /goal: start structuring a goal from the current request
• /goal template: show the blank five-field template
• "set a goal" / "define the goal" / "plan this properly": natural-language triggers that open the same flow

Example Usage

User: "make the scanner better"

That's a wish, not a goal, there's no finish line and no way to prove success. Structured:

🎯 Goal: Finish line defined.

TASK: Improve the site scanner's transport-security scoring
WHY: Users trust the score to flag insecure sites; today HTTP and HTTPS
     sites can score the same, which is misleading
OUTCOME: The scan assigns a strictly lower security score to a plain-HTTP
         site than to the same site served over HTTPS
CONSTRAINTS:
  - no new dependencies
  - don't change the score scale or the JSON output shape
VERIFICATION: run the scanner against http://example.test and
              https://example.test (same content) and confirm the HTTP
              scan's numeric score is lower than the HTTPS scan's

Why this matters: a vague version ("scanner feels better," "tests pass") could ship with the HTTP and HTTPS scores still identical and every existing test still green. The build passes while the behavior is wrong. The VERIFICATION line makes that failure impossible to miss: two scans, one comparison, a criterion you either meet or you don't. Define that line before execution, and "done" stops being a matter of opinion.

Inputs

• The user's raw request or task description
• Optional: a drafted goal in the five-field labeled format (for validate)
• Engine: python -m memstack_skill_loader.goal (template / validate)

Outputs

• A structured goal with all five fields, VERIFICATION concrete and checkable
• Validation results: errors (not ready) and warnings (tighten before trusting)
• A finish line that execution and review can both be measured against

Level History

• Lv.1: Base: Five-field goal structuring (TASK / WHY / OUTCOME / CONSTRAINTS / VERIFICATION) with a required, checkable VERIFICATION criterion; backed by the shared goal engine (goal_engine.py) used by both this skill and the Architect stage. (Origin: MemStack, Jul 2026)