Content Digest

Derived at runtime

• DEL-ID — from _CONTEXT.md or inferred from the deliverable folder name
• DEL Name — from _CONTEXT.md
• PKG-ID — from _CONTEXT.md

Optional

• None.

Runtime overrides

Read boundary

Reads are limited to files within the specified deliverable folder:

The skill must NOT read files outside the specified deliverable folder. It must NOT cross into adjacent deliverable folders or scan package/project-level registers.

Write boundary

Writes are limited to:

• {RuntimeOverrides.OUTPUT_PATH} — exactly one file per run

The parent directory of OUTPUT_PATH must already exist. The skill does not create the directory.

Tool usage

• No deterministic tools. This is a reasoning-first, semantic-matching extraction skill.
• The allowed-tools frontmatter field is intentionally omitted.

Disallowed behavior:

• No writing outside OUTPUT_PATH
• No modification of any file in DELIVERABLE_PATH
• No reading files outside the specified deliverable folder
• No cross-deliverable scanning or comparison

Method

Step 0 — Preconditions

1. Validate DELIVERABLE_PATH exists and is a directory. If not: report RUN_STATUS=FAILED_INPUTS.
2. Validate parent directory of OUTPUT_PATH exists. If not: report RUN_STATUS=FAILED_INPUTS.

Step 1 — Read deliverable files

Read each file listed in the Read Boundary table, respecting the specified read scope (full or line-limited). If a file is absent, record that fact; do not fail.

Step 2 — Extract structured information

From the files read, extract:

• Identity (from _CONTEXT.md): Package (PKG-ID and name), Discipline, Type, Responsible
• Scope (from _CONTEXT.md): Description (1-2 sentence summary), Acceptance Criteria (bulleted list), SOW Traceability IDs, Objective Traceability IDs
• Document Kit Summary (from first 80 lines of each kit document): one-sentence focus summary for Datasheet, Specification, Guidance, Procedure
• Dependencies (from _DEPENDENCIES.md): Tracking mode (NOT_TRACKED/DECLARED/TRACKED), upstream count, downstream count, key upstream (top 3-5 by name), key downstream (top 3-5 by name)
• References (from _REFERENCES.md): list of all referenced documents
• Semantic (from first 40 lines of _SEMANTIC.md): whether present (YES/NO), framework type if identifiable

Step 3 — Note quality observations

Scan the read content for and record specific items:

• TBD items and their resolution authority
• ASSUMPTION labels and their rationale
• Placeholder text or incomplete sections
• Conflict Table entries with HumanRuling status
• Enrichment markers (e.g., A-001, B-001)
• Inconsistencies between documents within the folder

Quality observations are factual flags, not evaluative judgments.

Step 4 — Write digest

Write the structured digest to OUTPUT_PATH using the 7-section structure defined below. The file is overwritten if it already exists.

Output structure

The digest file contains exactly seven sections, in this order:

## {DEL-ID} — {Name}

### Identity
- Package: {PKG-ID and name}
- Discipline: {from _CONTEXT}
- Type: {from _CONTEXT}
- Responsible: {from _CONTEXT}

### Scope
- Description: {1-2 sentence summary from _CONTEXT}
- Acceptance Criteria: {bulleted list from _CONTEXT}
- SOW Traceability: {SOW-IDs}
- Objective Traceability: {OBJ-IDs}

### Document Kit Summary
- Datasheet focus: {1 sentence}
- Specification focus: {1 sentence}
- Guidance focus: {1 sentence}
- Procedure focus: {1 sentence}

### Dependencies
- Tracking mode: {NOT_TRACKED/DECLARED/TRACKED}
- Upstream count: {N}
- Downstream count: {N}
- Key upstream: {top 3-5 by name}
- Key downstream: {top 3-5 by name}

### References
- {list all referenced documents}

### Semantic
- _SEMANTIC.md present: {YES/NO}
- Framework type: {if identifiable}

### Quality Observations
- {TBDs, placeholders, inconsistencies — specific items with IDs where available}

Missing information is noted as not present or TBD in source. Do not leave placeholder braces in the output.

Outputs

• One digest file at {RuntimeOverrides.OUTPUT_PATH}, containing the seven required sections.

Non-negotiable constraints

• Read-only on deliverable production documents. Never modify any file in DELIVERABLE_PATH.
• Single-deliverable scope. Each run processes exactly one deliverable folder. No cross-deliverable scanning.
• No invention. Report what the files contain. Do not infer, speculate, or fill gaps. Missing information is noted explicitly.
• Semantic matching, not deep reasoning. Extract structured information from document patterns. Flag quality observations but do not perform engineering judgment.
• One digest per run. Each invocation produces exactly one digest file at OUTPUT_PATH.
• Seven-section structure preserved. The output must contain all seven sections (Identity, Scope, Document Kit Summary, Dependencies, References, Semantic, Quality Observations) in the specified order.
• No engineering judgments. Evaluative assessment of deliverable correctness is EVALUATION_REPORT's responsibility, not this skill's.

QA expectations

• All seven sections present in the output digest.
• Identity fields populated from _CONTEXT.md (not invented).
• SOW and OBJ traceability IDs listed (or explicitly marked absent).
• Dependency counts stated with key upstream/downstream named.
• Quality Observations are specific items (TBDs, assumptions, conflicts) with IDs where available — not vague assessments.
• No files in DELIVERABLE_PATH were written or modified.
• The digest was written to the specified OUTPUT_PATH.
• No files outside the deliverable folder were read.

Downstream consumer

This skill produces one digest per deliverable. EVALUATION synthesizes digests across packages; EVALUATION_REPORT applies evaluative reasoning when scoring dimensions. This skill does not perform scoring or cross-deliverable synthesis.