Usage#

/spec:ambiguity                      # Evaluate the latest Seed spec
/spec:ambiguity --file {path}        # Evaluate a specific file
/spec:ambiguity --verbose            # Include detailed per-dimension analysis

Parameters#

Execution Flow#

1. Load Target File#

• Read Seed spec file
• Parse each section

2. 5-Dimension Evaluation#

Evaluate each dimension per AMBIGUITY_RUBRIC.md:

2a. Lexical Precision (weight 0.25)

• Scan for prohibited words in the spec's own language — English ("fast", "appropriate", "etc.", "if needed" ...) AND Korean ("빠르게", "적절히", "등등", "필요시" ...). Use the locale-tagged table in AMBIGUITY_RUBRIC.md.
• Use semantic judgment, not keyword-matching alone — a vague phrase in any language counts even if it is not in the table.
• If the spec language is not covered by the table, mark this dimension UNVERIFIED (not 0.0) and fall back to semantic judgment — never score an un-scannable language as "0 prohibited words".
• Check whether specific values/criteria have been substituted
• Score: 0.0 (0 prohibited words) ~ 1.0 (pervasively ambiguous)

2b. Reference Clarity (weight 0.20)

• Scan for pronouns/demonstratives — English ("this", "that", "said") AND Korean ("이것", "그것", "해당", "위의")
• Check whether abbreviations are defined
• Check for implicit subjects ("is processed" → missing agent)
• Score: 0.0 (all references clear) ~ 1.0 (key references unclear)

2c. Completeness (weight 0.20)

• Check whether AC (acceptance criteria) exist for all features
• Check whether edge cases are identified
• Check whether error scenarios are defined
• Score: 0.0 (all complete) ~ 1.0 (most missing)

2d. Consistency (weight 0.20)

• Check terminology unity
• Check for numeric/logical contradictions between sections
• Check priority consistency
• Score: 0.0 (fully consistent) ~ 1.0 (severe contradictions)

2e. Testability (weight 0.15)

• Check whether each requirement has Pass/Fail criteria
• Check automation feasibility
• Check reproducibility
• Score: 0.0 (all testable) ~ 1.0 (most require subjective judgment)

3. Final Score Calculation#

score = (lexical × 0.25) + (reference × 0.20) + (completeness × 0.20)
      + (consistency × 0.20) + (testability × 0.15)

4. Verdict#

5. Seed Spec Update#

• Update the ambiguity score in metadata

Output#

Evaluation results printed to console. With --verbose, detailed per-dimension analysis is included. Ambiguity score in Seed spec metadata is updated.

Related Commands#

• /spec:seed — Create Seed spec
• /spec:verify — Verification + lock
• /spec:evaluate — 3-stage integrated evaluation

References#

• references/AMBIGUITY_RUBRIC.md — Detailed 5-dimension rubric
• config/thresholds.yaml — Threshold settings