Your docs are lying
to your agents.
Written process rots the day it ships, and your agents follow the official story straight into failure. Lore reads how your company actually works, writes the cited Skills your agents run, and flags the moment reality drifts from the doc.
Your scattered sources
Sarah: refunds under $500 the CSM can just approve. above that loop in finance. and the annual grandfather thing still applies for anyone before Q1
“Anything over $5k needs CFO sign-off now. No exceptions after the March incident.”
all refunds require manager approval
One cited Skill agents run
When this Skill applies
A customer requests a refund or cancellation.
Decision criteria
Edge cases
Messy sources in · one cited SKILL.md out · loaded by Claude Code via MCP
Connects to the tools your team already lives in
GranolaThe difference
A skill layer, not a memory layer.
Memory tools store notes for your agent to read, then it improvises. Lore writes the cited playbook your agent runs. Closer to a compiler than a search engine.
Memory layers
Mem0 · Zep · Letta · Glean
- Store notes and hand them back on request
- The agent reads them, then improvises a response
- Relevance scores, not decisions. You never see which rule it used
Lore
The skill layer
- Writes the cited playbook the agent executes
- Every line links to the Slack message, PR, or email that set it
- Re-emits the moment a rule changes, so it is never stale
We use a memory layer underneath. We’re not one. The artifact your agent runs is the SKILL.md and that’s the part nobody else is building.
The audit loop
Every run, logged.
When an agent loads a Skill or runs it on a ticket, the call lands here. Version, agent, outcome, duration. Failed runs flag the Skill for the next refresh.
Runs . incident-response
Every time an agent loads or applies this Skill, one row lands here. Failed runs surface as candidates for Skill improvement.
| When | Agent | Kind | Ver | Dur | Status |
|---|---|---|---|---|---|
| just now | Claude Code | execute | v1.0.4 | 5.2s | successok |
| 3m ago | Cursor | load | v1.0.4 | 0.1s | successok |
| 12m ago | Lore dashboard | execute | v1.0.4 | 6.8s | successok |
| 1h ago | Claude Code | execute | v1.0.3 | 32.0s | failederr |
| 2h ago | Windsurf | load | v1.0.3 | 0.1s | successok |
| 3h ago | Claude Code | execute | v1.0.3 | 31.5s | failederr |
| 4h ago | Codex CLI | load | v1.0.3 | 0.1s | successok |
| 6h ago | Claude Code | execute | v1.0.3 | 5.9s | successok |
Connect your team.
Watch your agents stop guessing.
Be one of the ten teams whose agents run on real, cited process instead of guesswork. Connect Slack, GitHub, Linear, and email. Your first Skills land in an afternoon. We reply within 24 hours.