Built and audited agent context-routing site

2 min read

Built and audited agent context-routing site

  • Day: 2026-04-21
  • Time: 10:30 to 10:40
  • Project: Dev
  • Workspace: WP 2: Operational
  • Status: In Progress
  • Priority: HIGH
  • Assignee: Matías Nehuen Iglesias
  • Tags: Context-Routing, Source-Registry, Docusaurus, Agent-Usability, Static-Site, Knowledge-Management

Description

Session Goal

Advance a governed AI context-routing layer from concept to a usable publication surface, then validate whether the resulting registry/site is actually agent-consumable.

Key Activities

  • Defined the capture system as having crossed a threshold from raw intention into a governed intake layer with canonical vocabulary, routing logic, artifact types, and anti-drift constraints.
  • Designed a shared operational context spine around canonical objects such as activities, artifacts, entities, touches, snapshots, and open loops.
  • Reframed the memory problem as a thin integration layer over existing sources rather than full centralization, with a source registry, trust tiers, freshness semantics, and task-specific materialized views.
  • Drafted and refined a source registry v0 for MAL/context routing, including source IDs, locations, trust ordering, access modes, and likely questions.
  • Proposed a staged publication architecture: registry index, source wrapper pages, hosted artifacts/snapshots, and controlled exposure via static publishing.
  • Implemented/outlined a Google Sheet → Docusaurus routing surface generator, including validation, static asset handling, and site scaffold/documentation structure.
  • Triaged Docusaurus build issues caused by duplicate routes, missing static JSON assets, and broken artifact/snapshot links; identified pathname:/// as the correct pattern for static file linking.
  • Validated the live site and audited routing quality across project, job-search, manuals, and recent-work sources.

Achievements

  • The registry-to-site pipeline reached a usable milestone: the site is live, the homepage and routing directory work, and sources.[[json]] is being served.
  • The session clarified that the next bottleneck is not more capture, but a shared operational substrate and better exposure of high-value sources.
  • The audit surfaced a clear quality split: manuals are well wired and actionable, while project, job-search, and recent-activity surfaces remain too metadata-heavy for downstream agents.
  • Concrete low-effort fixes were identified: clickable origins, published snapshot URLs, redacted state pages, deeper manual links, and conditional rendering of missing artifacts.

Pending Tasks

  • Run a narrow agent-usability test against the live context-routing site using evidence-based retrieval tasks.
  • Patch remaining publication defects: duplicate route ownership, missing static assets, and broken links.
  • Expand source exposure for weak surfaces, especially project, job-search, and recent-work entries.
  • Continue refining the registry schema and publication pipeline so the context surface remains governed, discoverable, and low-drift.

Evidence

  • source_file=2026-04-21.sessions.jsonl, line_number=4, event_count=0, session_id=ebcb7c3d5f7dfdcef82943994ab6fa1c8787d7c21ded78609f385bb378d70a42
  • event_ids: []

Graph View

Backlinks