Writing index
Notes that return to a system.
Each essay starts in an existing README or accepted ADR and returns to the project that made the question concrete.
Topics
pii-redaction 1 · agent-systems 2 · benchmarking 1 · fail-closed 1 · privacy-engineering 1 · retrieval 2 · citations 1 · rag 1 · evaluation 1 · developer-tools 1 · sqlite 1 · fts5 1 · trust-boundaries 1 · local-first 1
Index
Evidence before atmosphere.
Claims stay bounded to public project documentation and linked evidence.
Reliable PII redaction for modern agents
Agents leak PII at every boundary. mastra-pii started by measuring the data — Hinglish chat and Indian identifiers that off-the-shelf redactors miss — then built a fail-closed processor layer around what the benchmarks proved.
Citation-first RAG needs an insufficient-context path
A citation-first demo earns trust by showing its sources, naming its infrastructure, and treating insufficient context as a valid result.
Separate the product surface from the benchmark harness
DocNeedle's README is persuasive because it treats normal CLI use and repo-specific evaluation as different things instead of collapsing them into one story.
Trusted ingress before more tools
Career Copilot shows why a single-owner agent benefits from tight trust boundaries before it grows new capabilities.