Overview
Produce a unified summary of all completed pipeline analysis reports. The core principle: read every report, synthesize a composite executive summary of the full persona, then summarize each report individually — the reader should understand the complete picture from the top section alone, but can drill into any specific analysis below.
When to Use
- All (or most) pipeline skills have completed and written reports to
docs/analysis/ - A final deliverable is needed that combines findings across all phases
- Stakeholder needs a single document covering the entire analysis without reading 25+ separate reports
- Pipeline orchestrator signals that synthesis is complete and summary is the final step
When NOT to use:
- Reports are still actively being generated (wait for pipeline completion)
- Only a single report needs review (just read that report directly)
- You need to produce the subagent instruction prompt (that is skill 25, not this skill)
Core Pattern
Read all reports in docs/analysis/ (01 through 25)
|
v
[Executive Summary] -- synthesize key findings across ALL reports into a narrative
|
v
[Per-Report Summaries] -- summarize each report's findings in 3-8 bullet points
|
v
[Pipeline Health] -- note any failed/missing/low-confidence reports
|
v
Write to docs/analysis/26-pipeline-summary.md
Implementation
Step 1: Inventory Available Reports
Scan docs/analysis/ for all report files. Classify each as:
- Present: File exists with substantive content
- Missing: File does not exist (skill was not run or failed)
- Stub: File exists but is under 500 bytes or contains error markers
from pathlib import Path
REPORT_DIR = Path("docs/analysis")
EXPECTED_REPORTS = {
"01": "csv-metadata-forensic",
"02": "tiered-processing-pipeline",
"02b": "content-hydration",
"03": "automated-orchestration",
"04": "taxonomic-interest-classification",
"05": "nmf-topic-modeling",
"06": "llm-relevance-scoring",
"07": "vader-sentiment-analysis",
"08": "weighted-engagement-scoring",
"09": "supplementary-engagement",
"10": "network-social-graph",
"11": "longitudinal-growth-curves",
"12": "temporal-circadian-patterns",
"13": "taxonomic-shift-detection",
"14": "mdpi-hypernetwork-archetype",
"15": "big-five-personality",
"16": "liwc-psycholinguistic",
"17": "cat-linguistic-style-matching",
"18": "stylometric-fingerprinting",
"19": "readability-lexical-diversity",
"20": "rhetorical-discourse-structure",
"21": "register-variation-code-switching",
"22": "speech-act-pragmatic",
"23": "archetype-assignment",
"24": "style-specification",
"25": "subagent-instruction",
}
Step 2: Read and Synthesize
For each present report, read the full content. Then produce two outputs:
A. Executive Summary (top of report) This is NOT a list of bullet points from each report stitched together. It is a synthesized narrative that answers:
- Who is this person, in behavioral and communicative terms?
- What is their assigned archetype and why?
- What are the 3-5 most distinctive traits of their writing voice?
- What are the key constraints for replicating their voice?
- What is the confidence level, given data quality and analysis coverage?
Write this as 2-4 paragraphs of flowing prose. The reader should understand the complete persona from this section alone.
B. Per-Report Summaries For each completed report, write a section with:
- The report title and skill number
- 3-8 bullet points of key findings (quantitative where possible)
- Any caveats or low-confidence flags from the original report
Group these by pipeline phase:
- Data Preparation (reports 01-03 + 02b)
- Content and Interest Analysis (reports 04-06)
- Sentiment and Engagement (reports 07-10)
- Temporal and Behavioral (reports 11-13)
- Psycholinguistic Profiling (reports 14-22)
- Classification and Synthesis (reports 23-25)
Step 3: Pipeline Health Section
At the end of the report, include:
- Table of all expected reports with status (present / missing / stub)
- Count of completed vs. total skills
- Any reports that flagged low-confidence or insufficient data
- Whether the final subagent instruction (skill 25) was produced
Report Template
Write to docs/analysis/26-pipeline-summary.md:
# Pipeline Summary Report
Generated: [timestamp]
Reports analyzed: [N] of [total expected]
Pipeline status: [COMPLETE / PARTIAL]
---
## Executive Summary
[2-4 paragraphs synthesizing the complete persona picture. Who is this person?
What defines their voice? What are the key constraints for replication?
What is the confidence level?]
---
## Phase 1: Data Preparation
### 01 — CSV Metadata Forensic Reconstruction
- [key finding]
- [key finding]
- ...
### 02 — Tiered Processing Pipeline
- [key finding]
- ...
### 02b — Content Hydration
- [key finding]
- ...
### 03 — Automated Orchestration
- [key finding]
- ...
## Phase 2: Content and Interest Analysis
### 04 — Taxonomic Interest Classification
- [key finding]
- ...
[... continue for all phases and reports ...]
---
## Pipeline Health
| # | Report | Status | Confidence |
|---|--------|--------|------------|
| 01 | CSV Metadata Forensic | COMPLETE | high |
| 02 | Tiered Processing | COMPLETE | high |
| ... | ... | ... | ... |
### Missing Reports
- [list any missing reports and their impact on the overall analysis]
### Low-Confidence Flags
- [list any reports that flagged data quality issues or low confidence]
### Final Deliverables
- Archetype assignment: [present/missing] — docs/analysis/23-archetype-assignment.md
- Style specification: [present/missing] — docs/analysis/24-style-specification.md
- Subagent instruction: [present/missing] — docs/analysis/25-subagent-instruction.md
Good Patterns
- Synthesize, don't concatenate: The executive summary should tell a story, not repeat bullet points. Connect findings across reports — e.g., "The high Openness score (Big Five) aligns with the polymathic interest classification, and both are reflected in the user's register variation across 14 distinct subreddit communities."
- Quantify everything: Prefer "TTR of 0.72 indicating high lexical diversity" over "diverse vocabulary."
- Flag uncertainty explicitly: If a report noted low confidence or insufficient data, carry that caveat into the summary. Do not present low-confidence findings as definitive.
- Respect the archetype: The archetype assignment (skill 23) is the organizing frame. Lead with it in the executive summary and show how other findings support or nuance it.
- Include the subagent prompt status: The pipeline's ultimate deliverable is the subagent instruction. State clearly whether it was produced and at what confidence level.
Anti-Patterns
| Mistake | Why It Fails | Fix |
|---|---|---|
| Listing bullet points from each report without synthesis | Reader still has to do the mental work of connecting findings | Write flowing prose in the executive summary that draws connections |
| Omitting missing reports from the summary | Reader doesn't know what's missing and may assume completeness | Always include the Pipeline Health table showing all expected reports |
| Parroting numbers without interpretation | "Flesch-Kincaid 12.3" means nothing without context | Always translate metrics to plain language: "writes at a 12th-grade level" |
| Presenting low-confidence findings as definitive | Misleads about analysis reliability | Prefix with confidence level or caveat |
| Writing the summary before reading ALL available reports | Early reports bias the synthesis | Read all reports first, take notes, then write the summary |
Boundaries
This skill SHOULD:
- Read all available reports in
docs/analysis/ - Produce a synthesized executive summary narrative
- Summarize each report individually with quantitative findings
- Report pipeline health and missing/failed analyses
- Write to
docs/analysis/26-pipeline-summary.md
This skill should NOT:
- Re-run any analysis or produce new analytical findings
- Modify upstream reports
- Substitute its own judgment for the archetype assignment
- Produce the subagent instruction prompt (that is skill 25)
- Claim completeness when reports are missing
Insufficient Data Handling
| Condition | Action |
|---|---|
| Fewer than 10 reports present | Write summary of available reports; flag as "preliminary — insufficient coverage for full synthesis" |
| No archetype assignment (report 23) | Note absence in executive summary; present available findings without archetype framing |
| No synthesis reports (23-25) | Summarize phases 1-5 findings only; note "synthesis phase not completed" |
| Reports present but mostly stubs/failures | List what exists; write executive summary as "limited analysis" with explicit gaps |
| All reports present and complete | Full summary with high-confidence executive synthesis |
Common Mistakes
| Mistake | Fix |
|---|---|
| Writing the executive summary first, then the per-report sections | Read all reports first, write per-report summaries, THEN synthesize the executive summary from your notes |
| Making the summary longer than the sum of its parts | Each per-report section should be 3-8 bullets. The executive summary should be 2-4 paragraphs. Total should be concise. |
| Not linking back to source reports | Include the report path (e.g., docs/analysis/15-big-five-personality.md) so readers can drill into detail |
| Ignoring cross-report contradictions | If Big Five says low Openness but taxonomy says polymathic, note the tension explicitly |