docs(spec-grill): add Behavior vs. Hard Constraint tiebreaker rule

[sungjunlee]

Dogfooding finding (2026-05-23, PR #110; res scoped 2026-05-31).

Several predicates can be written either as an Expected Behavior or as a Hard Constraint. Example: triage-grooming's read-only default can be framed positively (default invocation is advisory; mutation requires --apply) or negatively (never mutate GitHub without --apply). Both can pass the 3-axis test.

Without a tie-breaker, two spec-grill sessions on the same capability can classify the same intent differently, creating avoidable diff churn.

Scope

Update skills/spec-grill/SKILL.md and/or skills/spec-grill/references/capabilities.md with the classification rule.

Proposed rule

• Positive normal outcome -> Expected Behavior.
• Bright-line negation / anti-Goodhart guard -> Hard Constraint.
• If both are valid, prefer the Hard Constraint only when the negative form protects against an optimization shortcut or data-loss shortcut.
• If explicit user consent is valid, avoid the loophole phrase "never X unless asked"; write the consent condition as a Behavior instead.

Acceptance Criteria

• spec-grill docs include the Behavior vs. Hard Constraint tie-breaker.
• At least one example shows the same intent expressed both ways and explains the recommended placement.
• Wording uses current spec-grill / spec/capabilities.md names, not retired backlog-charter paths.

[coderabbitai]

🔗 Related PRs

#99 - docs(backlog-charter): add good/bad verifiable-predicate examples (#95) [merged]
#110 - docs: dogfood spec/capabilities.md for dev-backlog itself (#105) [merged]
#160 - [codex] split spec charter and grill skills [merged]

---

📝 Issue Planner

_{Check the box below or use the @coderabbitai plan command to generate an implementation plan and prompts that you can use with your favorite coding assistant.}

• Create Plan

---

🧪 Issue enrichment is currently in open beta.

You can configure auto-planning by selecting labels in the issue_enrichment configuration.

To disable automatic issue enrichment, add the following to your .coderabbit.yaml:

issue_enrichment:
  auto_enrich:
    enabled: false

💬 Have feedback or questions? Drop into our discord!

[sungjunlee]

2026-05-31 triage update: scope moved from the retired backlog-charter name to spec-grill. Target docs are skills/spec-grill/SKILL.md and/or skills/spec-grill/references/capabilities.md. Keep the original intent: reduce grill-session diff churn when the same predicate can be written as either an Expected Behavior or a Hard Constraint. The useful remaining delta is an explicit tie-breaker: positive normal outcome -> Expected Behavior; bright-line negation / anti-Goodhart guard -> Hard Constraint; when both are valid, prefer the Hard Constraint only when the negative form protects against an optimization shortcut.