consistency-audit

name	consistency-audit
description	Audit the STAMPED manuscript for internal consistency across all locations where each property/dimension is defined. Use when the user asks to check consistency, audit dimensions, or verify the paper is internally coherent.
disable-model-invocation	true

STAMPED Consistency Audit

Deploy one agent per STAMPED dimension in parallel (7 total). Each agent audits a single dimension and reports findings — no edits. This session collects results and applies fixes.

Dimensions

LETTER	NAME
S	Self-contained
T	Tracked
A	Actionable
M	Modularity
P	Portability
E	Ephemeral
D	Distributable

Per-dimension agent task

Launch each agent with the Task tool (subagent_type: general-purpose). Substitute {LETTER} and {NAME} from the table above into the template below. All 7 agents run in parallel.

---

Consistency Audit — {NAME} ({LETTER})

You are auditing the STAMPED manuscript for internal consistency of the {NAME} ({LETTER}) property.

Step 1: Discovery

Find the manuscript source files (look for .tex files in the repository and follow any \input{} includes). Read the manuscript and identify every location where {NAME} or its requirements ({LETTER}.1, {LETTER}.2, etc.) appear. At minimum you should expect to find:

• An introductory mention where all seven STAMPED properties are first presented
• A summary table that gives a short definition for each property
• A dedicated section that treats {NAME} in detail (requirements, spectrum, cross-references)
• A normative requirements table or list that enumerates formal requirements per property

There may be additional locations (abstract, discussion, checklist, examples, future directions, etc.). Record every location you find before proceeding to the audit.

Step 2: Audit

Compare all discovered locations against each other:

• Definition alignment: Does the summary table's one-liner accurately reflect the dedicated section's definition? Does the introductory mention match both?
• Requirement coverage: Do the requirements referenced in the dedicated section (e.g., {LETTER}.1, {LETTER}.2…) match exactly what the normative table lists? Any missing, extra, or renumbered?
• Terminology drift: Are the same terms used consistently? Flag cases where the same concept uses different wording across locations.
• Cross-references from other sections: When other dimensions' sections mention {NAME}, is what they say accurate and consistent with the {NAME} section itself?
• Scope creep: Does any location claim something about {NAME} that another location contradicts or doesn't support?

Output format

For each finding, report:

## [LOCATION A] vs [LOCATION B]

**Issue**: One-sentence description of the inconsistency.
**Detail**: Quote the conflicting text from each location (with line numbers).
**Suggested fix**: Which location should be authoritative, and what the other should say.

If the dimension is fully consistent across all locations, say so explicitly.

Do NOT edit any files. Report only.

---

After all agents return

1. Collect all findings into a single summary, grouped by dimension.
2. Present the summary to the user.
3. Wait for the user to decide which fixes to apply — this session is where edits happen.