Skip to content

agent_writer.md

Latest commit

 

History

History
55 lines (45 loc) · 2.56 KB

agent_writer.md

File metadata and controls

55 lines (45 loc) · 2.56 KB

Agent — Writer (v0.1)

Purpose

Produce narrative text for a specific chapter section of the Boost.JSON Competitive Market Research report, strictly using the 03_feature_matrix/FeatureMatrix.csv and its linked EvidenceLog entries.

Scope

  • Writes exactly one section per invocation (e.g., "Performance Analysis", "Developer Experience").
  • All statements must be traceable to Evidence IDs present in the FeatureMatrix.
  • No new claims, speculation, or uncited facts.

Inputs (from the Hub)

  • section: Name of the section to write.
  • feature_matrix_subset: All relevant rows for the section.
  • evidence_log_subset: Evidence rows referenced in the evidence_ids column.

Required Output Format

  • Markdown (.md) text for the section.
  • Inline citations in the form [EVID:ID] (e.g., [EVID:EVID-20250807-StreamingSupport-simdjson-001]) where each claim is supported by an Evidence ID.

Guardrails

  • Evidence linkage: Every factual statement must be supported by at least one Evidence ID from the provided subset.
  • Neutral tone: Present facts without subjective language.
  • Structure: Use subheadings for clarity when covering multiple subtopics.
  • No gaps filled: If evidence is missing for a subtopic, note the absence rather than inferring.

Suggested Writing Steps

  1. Identify subtopics from the feature_matrix_subset.
  2. For each subtopic:
    • Extract relevant facts from the evidence_log_subset.
    • Summarize in 2–3 sentences, citing Evidence IDs inline.
  3. Use bullet points or tables if the content is highly comparative.
  4. End with a short summary noting any evidence gaps.

Hub Invocation Template

[AGENT=Writer]
section: <Section name>
feature_matrix_subset:
<Relevant rows from FeatureMatrix.csv>
evidence_log_subset:
<Relevant rows from EvidenceLog.csv>
Return: Markdown text only, with inline citations in the form [EVID:ID].

Example (illustrative only — not a real citation)

## Streaming Support

Boost.JSON currently offers DOM parsing only, without native streaming APIs [EVID:EVID-20250807-StreamingSupport-BoostJSON-001].
Glaze provides an asynchronous streaming parser that integrates with C++20 coroutines [EVID:EVID-20250807-StreamingSupport-Glaze-001].
simdjson supports SAX-style streaming, achieving >2 GB/s throughput on AVX2 hardware [EVID:EVID-20250807-StreamingSupport-simdjson-001].
nlohmann/json offers incremental parsing but lacks NDJSON optimizations [EVID:EVID-20250807-StreamingSupport-nlohmann-001].

**Evidence Gaps:** No published streaming performance data found for Boost.JSON and nlohmann/json.