Entry-Point Prompt + Personas Obligation

Dieser Inhalt ist noch nicht in deiner Sprache verfügbar.

Spec: gradingSpec/1.1.0 Status: stable (additive extension of 1.0.0) Changes vs. 1.0.0: entirely new Entry-Point Prompt chapter (entry-point prompt template + personas obligation for Single AND Selection).

Every grading run begins from an empty context, and an empty context needs a concrete starting signal. This chapter defines that signal: the entry-point prompt, the first prompt a grader pastes after /clear, which binds the persona, the selection or single schema, the mode, and the spec version into one block. It also fixes the personas obligation — when a persona is spec-mandatory versus merely conventional — so that Single- and Selection-Gradings stay anchored to the same evaluation lens.

Entry-Point Prompt

Empty context (see 02-eligibility.md) is a convention — it needs an operational anchor. The entry-point prompt is that anchor: the first prompt after /clear starts every grading run and binds persona, selection/schema, mode, and spec version.

Concept

Term	Definition	Checkable?
Empty context	Convention: no relevant prior information	No, trust-based
Entry-point prompt	First prompt after `/clear` = grading start	No, organisational

The spec requires: the README in the grading repository provides a prompt template. The grader runs /clear, copies the prompt, and fills in persona, selection/single-schema, mode, and the lockSnapshot hash from the selection’s index.json.

Prompt Template

The binding prompt template is:

You are performing a FlowMCP grading. Instructions:

1. Persona: crypto-trader-2026
2. Selection: crypto-domain-full, lockSnapshot hash: <sha>
3. Mode: Full (initial baseline)
4. Spec version: gradingSpec/3.0.0
5. Pre-condition: all member schemas have gradingStatus: stable
6. Output format: _gradings/<area>--<timestamp>.json

Six numbered lines. For Single-Gradings, line 2 is replaced and line 5 is dropped:

1. Persona: crypto-trader-2026
2. Single-Schema: etherscan.getContractEthereum, area: single-test
3. Mode: Full (initial baseline)
4. Spec version: gradingSpec/3.0.0
6. Output format: _gradings/<area>--<timestamp>.json

(Line 5 is dropped — the pre-condition applies only to aggregated checks, see 21-pre-conditions.md.)

Personas Obligation

The personas obligation has two levels — spec obligation and convention. They are complementary (not alternative).

Level	Applies to	Anchoring
Spec obligation	Selection only	`personaIds[]` is a mandatory field in `selection.json` (see `16-selection-lockfile.md`)
Convention via prompt	Single AND Selection	Every grading run starts with the persona line in the prompt (line 1)

Rationale: Single-Gradings without a persona anchor produce evaluation drift. The spec obligation in selection.json is the formal anchoring; the prompt requirement is the operational one.

Personas-obligation summary:

Selection-Gradings: persona MUST be in selection.json AND in the prompt
Single-Gradings: persona SHOULD be in the prompt (organisational, not a spec-mandatory field)

Where the Template Lives

The README in the grading repository carries the ready-to-use prompt with example values, anchoring the entry-point prompt and the personas obligation in a single block. The grader copies that block, runs /clear, and fills in the run-specific values (persona, selection or single schema, mode, and the lockSnapshot hash).

./00-overview.md — see chapter 00.
./02-eligibility.md — see chapter 02.
./12-personas-contract.md — see chapter 12.
./16-selection-lockfile.md — see chapter 16.
./19-folder-layout.md — see chapter 19.
./21-pre-conditions.md — see chapter 21.

Entry-Point Prompt + Personas Obligation