Docs thatstay true to your Salesforce org.
LivingContext closes the loop between every commit, every metadata change, and the runbooks they make wrong — proposing the anchor-precise edits so your docs always describe the system you actually have.
How to approve an opportunity over $250 000
Opportunities over $250 000 route to the Sales-Director Regional-VP for approval. Approval uses single-approver two-of-three quorum as of approval process v3.1.
Required fields: Stage, CloseDate, ContractTerm__c.
{ "anchor": "Sales-Director",
"new": "Regional-VP" }Six surfaces. One loop. No quarterly doc audits.
Each surface is independently usable. Adopt one — the changelog agent, the MCP server, the tag panel — and add the others as your team is ready.
From the file you just changed, to the runbooks that mention it.
Two-tier retrieval — graph traversal first, change-driven fallback second. One mapper handles every entity type: Apex, Flows, Fields, Validation Rules, Approval Processes.
Five labels. Five queues. Less noise.
Reviewers triage by intent, not by alert fatigue. Per-workspace calibration on false-positives.
Your AI assistants stop guessing.
Claude Desktop, Cursor, Zed — point them at the MCP server and they only see docs known to be in sync with the metadata they describe.
The model can't rewrite your doc. It can only point and surge.
Anchor-based ops { op, anchor, new }. Reviewers triage each edit independently — a noisy fifth edit doesn't block the four good ones.
{ op: "replace",
anchor: "Sales-Director",
new: "Regional-VP" }What's documented. What isn't.
Per-entity coverage at a glance. The next-action queue is driven by data, not who shouted loudest in #sf-platform.
Your repo is the canonical store.
Confluence and SF Knowledge stay editable. Git carries history, diff, blame, rollback — portable even if you leave.
SF-flavored release notes, on every tag.
Built on git-cliff. Groups commits by SF entity, not conventional-commit type. Drop the Action in your repo.
uses: livingcontext/changelog-action@v0
with:
range: v1.4.0..v1.5.0
enrich: true
From the commit
to the corrected doc.
Continuously.
No quarterly audits. No “who-shouted-loudest” queue. Every change is either accounted for, surfaced as a known false-positive, or already on its way to publish.
Every commit and metadata change is observed in near-real-time.
Two-tier retrieval walks the doc-code graph, then change-driven fallback.
Each affected doc receives a typed staleness verdict.
{ op: "replace",
anchor: "Sales-Director",
new: "Regional-VP" }Anchor-based edits — the model surges existing strings, never rewrites.
Reviewers approve or reject each edit; the queue waits for the doc.
Confluence / SF Knowledge updated; every change mirrored to git.
“Other tools detect doc drift. We close the loop — every change is either resolved or it stays in the queue. Forever.”
An engineer ships a metadata change.
An Apex class is renamed, an approval process gains a quorum step, a validation rule expands. Somewhere — in a runbook nobody opened this quarter — a sentence just became wrong.
We find the docs that just turned false.
Two-tier retrieval: the graph names exact candidates from prior tagging; change-driven fallback catches anything inferred from the diff text. One mapper, every entity type.
Each doc earns a typed verdict.
NowWrong, NowIncomplete, NowMisleading, StaleButCorrect, FalsePositive. Reviewers triage by intent — not by alert fatigue or a quarterly audit reminder.
Confluence updates. Git remembers.
The reviewer approves the edits they want. Confluence and SF Knowledge update; the customer's own git repo records every change forever — diff, blame, rollback intact.
<approvalStep> <name>Director Approval</name> <assignedApprover> - <approver>Sales-Director</approver> + <approver>Regional-VP</approver> </assignedApprover> - <approverType>singleApprover</approverType> + <approverType>quorum</approverType> + <quorum>2-of-3</quorum> </approvalStep>
From commit to corrected doc.
An engineer ships a metadata change.
An Apex class is renamed, an approval process gains a quorum step, a validation rule expands. Somewhere — in a runbook nobody opened this quarter — a sentence just became wrong.
<approvalStep> <name>Director Approval</name> <assignedApprover> - <approver>Sales-Director</approver> + <approver>Regional-VP</approver> </assignedApprover> - <approverType>singleApprover</approverType> + <approverType>quorum</approverType> + <quorum>2-of-3</quorum> </approvalStep>
We find the docs that just turned false.
Two-tier retrieval: the graph names exact candidates from prior tagging; change-driven fallback catches anything inferred from the diff text. One mapper, every entity type.
Each doc earns a typed verdict.
NowWrong, NowIncomplete, NowMisleading, StaleButCorrect, FalsePositive. Reviewers triage by intent — not by alert fatigue or a quarterly audit reminder.
Confluence updates. Git remembers.
The reviewer approves the edits they want. Confluence and SF Knowledge update; the customer's own git repo records every change forever — diff, blame, rollback intact.
# git log --oneline docs/ 4d9a2c3 lc: approve edit 2/2 on opp-approval.md 7e4f2b1 lc: approve edit 1/2 on opp-approval.md
NowWrong / NowIncomplete / NowMisleading / StaleButCorrect / FalsePositive
After per-workspace calibration on real reviewer behavior.
Onboarding wizard scans + infers + calibrates in one sitting.
Confluence and SF Knowledge stay as editing surfaces. Git is the mirror.
Release notes that understand Salesforce.
Built on git-cliff. Groups commits by Salesforce entity — Apex, Custom Objects, Custom Fields, Flows, Validation Rules, Approval Processes — not by raw conventional-commit type. Drop the GitHub Action into a workflow; get notes back as a PR comment, a build artifact, or a release-page body.
For assistants that shouldn't guess.
Point Claude Desktop, Cursor, Zed, the IDE agent of the week at the LivingContext MCP server. They only see docs known to be in sync with the metadata they describe — no confident hallucinations from a runbook that was true in March.
Docs that ship as clean Markdown. With a file just for the machines.
Your repo is the source of truth, so docs ship as Markdown natively — no scraping a rendered DOM, no fighting a CMS export. At every publish, LivingContext writes a /llms.txt index in the emerging convention, so Claude, Cursor, Perplexity, ChatGPT — and the next agent that hasn't shipped yet — can read your verified docs with zero integration work.
# LivingContext for Acme RevOps > Verified Salesforce docs for the Acme RevOps org.
> All entries are in-sync with metadata as of 7e4f2b1. ## Runbooks - [Opportunity approval over $250k](/docs/runbooks/opp-approval.md): How to approve opps over $250k. Routes to Regional-VP w/ 2-of-3 quorum. - [Discount policy](/docs/runbooks/discount-policy.md): Approval matrix and routing rules by discount tier. - [Case auto-close](/docs/runbooks/case-auto-close.md): Conditions for automatic case closure. ## Reference - [Custom fields glossary](/docs/reference/fields.md): All custom fields with type, description, validation. - [Approval processes](/docs/reference/approvals.md): Active approval processes by entity. ## Optional - [Legacy approval flow](/docs/legacy/old-approval.md): Removed Apr 2026. Kept for audit only.
Other tools manage docs.
We make them true again.
Stop trusting docs.
Start verifying them.
Drop your email. We'll run a calibration pass on your repo and send back a draft coverage report within seven days.