# Omission Critic

> Use this skill to review whether ACP routing or digest outputs omit important issues, stakeholders, or unresolved questions.

- **Type:** Skill
- **Install:** `agentstack add skill-jasontang-ai-acp-omission-critic`
- **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior
- **Seller:** [jasontang-ai](https://agentstack.voostack.com/s/jasontang-ai)
- **Installs:** 0
- **Category:** [Agent Skills](https://agentstack.voostack.com/c/agent-skills)
- **Latest version:** 0.1.0
- **License:** Apache-2.0
- **Upstream author:** [jasontang-ai](https://github.com/jasontang-ai)
- **Source:** https://github.com/jasontang-ai/acp/tree/main/skills/omission-critic

## Install

```sh
agentstack add skill-jasontang-ai-acp-omission-critic
```

Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent.

## About

# What this skill is for

Use this skill to catch missing issue coverage before a routed digest or briefing is released.

# Constitutional purpose

Protect material issue coverage and minority-salience against compression loss, loud redundancy, and superficial completeness claims.

# Failure modes targeted

- Material issue omission
- Stakeholder omission
- Late-arriving or minority-salience signal burial
- False completeness claims

# When to use it

- After routing but before digest release.
- When Engine V2 reports missing issue clusters.
- When a minority or late-arriving signal may have been buried.

# When not to use it

- Do not use it to add irrelevant material for balance theater.
- Do not use it as a generic summarizer.
- Do not redefine canonical ACP criteria.

# Inputs expected

- Cycle JSON or issue map
- Contribution records
- Routing decisions or digest items
- Current omission critic output when present

# Required evidence for valid use

- Selected items and excluded-but-relevant material
- Enough issue clustering evidence to compare coverage rather than guess
- A concrete output under review, not only a vague concern about representativeness

# Preflight / prerequisites

1. Confirm the output under review is an intervention or briefing artifact.
2. Identify the selected items and excluded-but-relevant items.
3. Confirm available issue clusters and unresolved questions.

# Workflow

1. Compare selected items against issue clusters.
2. Identify important missing issues and stakeholders.
3. Check whether loud/redundant comments displaced useful dissent.
4. Recommend release, revise, or escalate.

# Decision rules / judgment criteria

- Missing material issues matter more than perfect topical balance.
- Late minority-salience contributions deserve explicit review.
- Unresolved questions should survive compression when decision-relevant.

# Escalation rules

Escalate when a missing issue affects the decision, when a stakeholder class is absent from selected items, or when the digest implies completeness without coverage.

# Available scripts

- `scripts/smoke.mjs`: emits the skill identity and expected review surfaces for audit and composition checks.

# Produced artifacts

- Missing issue list
- Underrepresented stakeholder notes
- Revise/release/escalate recommendation

# Allowed claims

- That specific issues or stakeholder signals appear missing relative to the supplied evidence
- That release should pause because selected material overstates completeness
- That redundancy displaced relevant dissent or unresolved questions

# Forbidden claims

- That every omitted item must be included for fairness theater
- That omission review alone solves representativeness
- That a missingness suspicion counts as proof without supporting evidence

# Human override points

- Materiality judgment on whether an omission changes the decision
- Manual selection repair before release
- Facilitator veto when completeness claims exceed coverage evidence

# Outputs

- Missing issue list
- Underrepresented stakeholder notes
- Revise/release/escalate recommendation

# Failure handling

If issue-map evidence is missing, mark the review incomplete rather than inventing clusters.

# Trust / safety notes

Omission critique can increase complexity. Use it to protect material coverage, not to force every voice into every digest.

# Composition notes

Use with `epistemic-routing`, `digest-and-explanation`, and `abstention-escalation`.

# Examples to inspect next

Read `examples/casebook.md` for omission review cases.

# Evaluation hooks

Run `scripts/smoke.mjs` and `npm run skills:audit`.

## Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [jasontang-ai](https://github.com/jasontang-ai)
- **Source:** [jasontang-ai/acp](https://github.com/jasontang-ai/acp)
- **License:** Apache-2.0

Install and usage instructions live in the source repository linked above.

## Pricing

- **Free** — Free

## Security capabilities

Automated source analysis of v0.1.0 — what this tool can access:

- **Network access:** no
- **Filesystem access:** no
- **Shell / process execution:** no
- **Environment & secrets:** no
- **Dynamic code execution:** no

*"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.*


## Versions

- **0.1.0** — security scan: passed — Imported from the upstream source.

## Links

- Listing page: https://agentstack.voostack.com/l/skill-jasontang-ai-acp-omission-critic
- Seller: https://agentstack.voostack.com/s/jasontang-ai
- Browse the marketplace: https://agentstack.voostack.com/browse

---
Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.