Signal
Source knowledgeContinuous read of Slack, docs, tickets, and calls. Every fragment is timestamped, attributed, and hashed.
SkillLayer listens to your Slack, docs, tickets, and decisions, extracts the domain knowledge that actually runs the business, and compiles it into evidence-backed, human-approved Skills Files ready for any agent.
Request accessmessages observed daily
skills compiled, version-locked
evidence-traceable outputs
median draft to approved
SkillLayer connects through read-only adapters to the systems where decisions are actually made. It indexes message-by-message and document-by-document, never moving data or training third-party models.
Each stage advances knowledge toward something an agent can act on safely. No stage runs without the previous one signing off. The pipeline is auditable end to end.
Continuous read of Slack, docs, tickets, and calls. Every fragment is timestamped, attributed, and hashed.
Recurring patterns become candidate facts. Each cites three or more independent sources before promotion.
An executable procedure with preconditions, steps, escalation paths, and refusal rules, drafted but never published.
A named operator reviews evidence beside the draft. Comments, redlines, and approvals are versioned.
Signed, version-locked artifact. Drops into Claude, OpenAI, internal agents: same interface, same guarantees.
Open, declarative, signed. A Skills File is a single human-readable document that an agent loads at runtime, and that your compliance team can read on a Tuesday.
---
name: refund_legacy_account
version: 1.4.0
approved_by: maria.okonkwo@acme.com
approved_at: 2026-04-22T14:08Z
signature: 0x8a4f...c12e
---
# Skill: Refund a legacy-plan account
## Preconditions
- account.plan in { "starter-2019", "team-2019" }
- refund_amount <= $10,000 USD
- requester_role in { "csm", "support_lead" }
## Procedure
1. Verify account is in good standing.
2. if amount > $2,500 require finance.review()
3. Open Stripe refund, full or partial.
4. Annotate Salesforce: reason code, evidence link.
5. Notify customer using template T-217.
## Evidence
- slack://#ops-runbook/p1714382100
- notion://customer-tiers#refunds v.18
- linear://OPS-2204 decision
## Refusal
- if requester is unverified: halt, page #ops-oncall
- if amount > $10,000: halt, escalate to CFO
No skill ships without a named operator's signature. Reviewers see the draft, the evidence, and the diff against the previous version, side-by-side.
Add CFO escalation above $10K.
Tighten preconditions; pin template T-217.
First public release. Cited 3 sources.
Refused: missing finance handoff.
Skills Files are model-agnostic. The same compiled, approved skill loads into Claude, OpenAI, Vertex, or your internal agent and produces auditable behavior on the other side.
import { SkillLayer } from "@skilllayer/sdk";
const layer = new SkillLayer({ org: "acme", env: "prod" });
const refund = await layer.skill("refund_legacy_account@^1.4");
agent.use(refund, { policy: "strict", telemetry: true });
We retired four runbooks the day SkillLayer went live. Our agents stopped guessing.
Audit committee asked where an agent decision came from. We sent them a Skills File. That was it.
The version history is the part that sold us. We can prove what we knew and when.
Pilot in two weeks. Read-only adapters, your evidence, your approvers. We compile the first ten skills with you, in person.
Request access