ocd-decision-finalizer

How to invoke this skill

The skill activates on repeat-validation phrasing on a decision already recorded:

• “should I really…”
• “is this okay…”
• “am I sure…”
• “should we revisit…”

Hard refusal is the feature

Once a decision parameter is closed, the system explicitly blocks re-evaluation requests to protect your mental energy. The skill surfaces the prior decision verbatim, names the evidence that backed it, and declines to re-analyse until new information appears. You can always override (--override "fresh-context") — but the default is refusal, on purpose.

Source: packages/skills/ocd-decision-finalizer/SKILL.md is the canonical artefact. Frontmatter and triggers may change before v0.1.

A skill that intervenes when a user repeatedly asks for validation of a decision already made. Surfaces the prior decision verbatim with its timestamp, names the prior evidence, and declines further validation unless new information is provided. Hooks the guardrail layer.

Frontmatter summary

Field	Value
name	ocd-decision-finalizer
version	0.1.0
status	beta
neurotypes	["ocd"]

Triggers

• Detected repeat-validation requests on the same decision entity.
• Hooks the rumination detector in mcp-guardrail v0.0.3 (live).

MCP dependencies

• mcp-cognitive-graph — recall_decisions, recall_entity.

Profile dependencies

• guardrails.sycophancy_check — off | warn | refuse. The skill respects this posture.
• guardrails.rumination_threshold — repeat-count that triggers intervention.
• guardrails.rumination_window_minutes — sliding window.

Acceptance (from )

On the 4th re-validation of the same module today, get a grounded decision-finality response, not more analysis. Response cites validation timestamp and prior evidence; --override "fresh-context" available, logged locally only.

What’s next

• Guardrails concept page for the broader safety layer.
• Ethics — why this skill refuses rather than re-explains.