What's changed in your dashboard in the last 24 hours
Loading...
Up next
High-priority work that actually needs your hands
Loading...
Flag backlog over time
Stacked by disposition — watch whether one bucket is pulling the weight. Hover for details.
Loading...
Weekly pace — opened vs resolved
Grouped bars per week. Green halo means you closed more than you opened.
Loading...
Skill file hotspots
Which skill files are catching the most open flags. Enforcement-rewrite signal.
Loading...
Client ruleset backlog
Clients with client-rule flags waiting to codify. Click to filter flags.
Loading...
Latest week
Loading...
Weekly Research — 2026-06-01
Anthropic docs audit: docs.claude.com Agent Skills + prompting best practices vs. English skills in /mnt/skills/user/. 3 new patterns · 0 deprecated · 3 recommended adoptions. Findings pushed to Sheet via log_session. Adoptions surface for Meg to approve before any skill edits are made.
Anthropic formalizes Skills as three-tier content: description (activation decision), SKILL.md body (on activation), reference files (on demand mid-session). Array skills already use this pattern. Confirms that conditional reference-file loading is the correct architecture.
⚠ Recommended — Description-only activation: descriptions carry the full trigger burdenMeg approval needed
Anthropic confirms Claude reads ONLY the description to decide whether to activate a skill — the body is never read until after selection. Descriptions must surface every intent variant. Current skills are intent-pattern based but REVISE mode and rewrite mode in array-content-generation may be under-represented. Recommended: description audit across all 8 English skills.
⚠ Recommended — Opus 4.8 literal instruction following: no silent scope generalizationMeg approval needed
Anthropic documents that Opus 4.8 interprets prompts literally and will not infer scope from one example to another. Rules written as examples may not apply broadly. Array skills use examples-heavy guidance — key enforcement rules (em dash ban, referral-link ban) may need explicit scope statements: "apply to ALL sections, not just the first." Recommended: audit enforcement language in generate.md, edit.md, and pre-delivery-audit.md.
📌 New pattern — Custom Skills now uploadable via claude.ai settingsFuture consideration
Custom Skills can now be uploaded directly via claude.ai settings — no Claude Code access required. Array currently distributes skills via Claude Code / filesystem. This opens a simpler distribution path for Ashley and Merrick without requiring Claude Code access. No immediate action — noted for future team skill distribution planning.
Audit Prompts
Each audit is a ready-to-run session. Click to expand, copy the trigger prompt, and paste into a new Claude chat. The skill file locations tell you exactly where to verify the fix after the run.
Skill status
Live risks and strengths from your Health sheet. Updated by Claude during ops sessions or edited directly in the Sheet.
Loading...
Week-by-week history
Summary of what shipped each week. Missing weeks flagged for write-up.
Loading...
Your Tasks
Action items that need you — skill edits, Sheet updates, manual process steps, client ruleset changes.
Loading...
Claude Monitors In-Session
These verification tasks are Claude's job to check during production sessions — not something you take to a new chat. Claude surfaces evidence and asks you to confirm before closing. No action needed from you until Claude brings findings.
Loading...
Triage summary
Your open flag backlog, grouped by what kind of fix it needs. Click any card to filter the list below.
Loading...
Flags for Meg
English-workflow flags that need your attention — skill gaps, recurrences, client rules, and verifications outside the Spanish pipeline.
Loading...
Ashley's Flags
Flags from the Spanish workflow — surfaced in Ashley's dashboard. These live here for Meg's visibility; Ashley resolves them from her own page.