Maya Chen@maya_builds1d
A practical multi-agent review pattern that cut our evaluation cycles by 65%.
A planner, specialist reviewers, and a judge work from one structured rubric.
7:32 AM · Jul 12, 2026 · 18.4K Views
31 208 1.4K 18.4K
Your X bookmarks, compiled daily
Turn the useful workflowsyou save on X intoskills your agentscan use.
XSkills filters your bookmarks, researches the workflows worth keeping, and delivers evidence-backed skill drafts for you to approve in Telegram or the dashboard.
Trust and compatibility
Read-only bookmarksNo posting accessNothing installs without approval
Works with
- Claude Code
- Codex
- Cursor
- X bookmark.
- Verified by XSkills.
- You approve the exact package version.
- The skill is installed and ready for your selected agent.
01 X bookmark
02 XSkills verification
VerificationTelegram or dashboard
Verified by XSkills
Multi-agent review pattern
Research-backed workflow.
- Evidence
- 4 sources
- Quality
- 94/100
- Risk
- Low
Destination Codex
03 Your approval
Exact version · 8d73…c41a04 Agent-ready skill
multi-agent-review/v1.0.0
Installed in Codex
- SKILL.mdWhen to use, workflow, guardrails.
- references/Source post, research, citations.
- agents/openai.yamlCodex invocation defaults.
- eval_queries.jsonPositive and negative triggers.
checksum · 8d73…c41a
~/.agents/skills/multi-agent-review/
Saved in passing.Built with evidence.Approved by you.
The daily loop
Your day stays in Telegram.
The web app is for setup, recovery, inspection, and history. The recurring decision is one calm message.
01
Save naturally.
Bookmark useful AI workflows on X the same way you already do. No new inbox to maintain.
02
Review once in Telegram.
Get one researched batch in the conversation where your daily decisions already happen.
03
Approve what earns a place.
Install only the exact, checksum-bound skill package you have chosen to trust.
A specimen, followed through
From a saved post to a skill you can trust.
XSkills separates what the source proves from what still needs research, then packages only the useful, supportable workflow.
See the safety ledger01 · Source
Maya Chen@maya_builds1d
A practical multi-agent review pattern that cut our evaluation cycles by 65%.
A planner, specialist reviewers, and a judge work from one structured rubric.
7:32 AM · Jul 12, 2026 · 18.4K Views
31 208 1.4K 18.4K
02 · Evidence
Research ledger
- Core workflow extracted
- Four primary sources checked
- Constraints recorded
- Side effects graded low
03 · Package
multi-agent-review/v1.0.0
- SKILL.mdWhen to use, workflow, guardrails.
- references/Source post, research, citations.
- agents/openai.yamlCodex invocation defaults.
- eval_queries.jsonPositive and negative triggers.
checksum · 8d73…c41a
The safety ledger
Nothing crosses the line without you.
Automation prepares the work. A person still decides which exact package becomes agent capability.
Read-only by design
XSkills reads bookmarks. It never asks for permission to publish, reply, or alter your account.
Evidence stays attached
Sources, constraints, quality grades, and review notes travel with every draft.
Approval means this version
Your decision is bound to the package checksum. A revision always requires a new approval.
Your agents learn what you curate
Saved in passing. Built with evidence. Approved by you.
Connect X once. Review each draft where you work best. Let the useful things you notice become durable capability.
Connect