Syncolab Living Docs

Syncolab Living Docs (SKS)

Purpose

Use the Syngles Knowledge System (SKS) as the source of truth for Syncolab platform documentation: search before answering platform questions, fetch full docs by stable id, and propose patches for human review—never edit live docs directly.

When to Use

• Before answering about Syncolab platform behavior, schemas, integrations, or runbooks.
• Internal flows (CML, SKS, planning, work-sync, agent runtime) when unsure—search with the right facet.
• Proposing doc fixes when content drifts from code (minimal patch via sks_docs_propose_patch).

When NOT to Use

• Tenant agent memory (CML) → syncolab-cognitive-memory.
• Arbitrary vector indexes → vectorize.
• Direct live doc edits without the patch/review flow.

Expected Outcome

• Answers cite SKS doc ids used (e.g. [sks: platform.sks.architecture]).
• Search returns lightweight summaries; full body from sks_docs_get.
• Patches return contribution id; not applied until human approval.

Inputs to Gather

• Focused query for search.
• facet: internal, external, or both (default both).
• Optional scope, domain, loadTier (always | on_demand), limit.
• For patches: docId, one-line summary, minimal bodyPatch, optional reasoning.

Workflow

1. Search first for platform questions: sks_docs_search with focused query and correct facet.
2. Get full doc when needed: sks_docs_get with stable id from search.
3. Cite doc id in the reply.
4. If content is wrong: sks_docs_propose_patch with smallest diff; never direct live edit.
5. External runtimes: use hydrated sks_layer endpoints with Bearer JWT same as tool semantics.

Tools

• sks_docs_search — query (required), facet, scope, domain, loadTier, limit. Returns id, title, summary, scope, domain, authority, updatedAt.
• sks_docs_get — stable id; full markdown, headings, references, source binding, load tier.
• sks_docs_propose_patch — docId, summary, bodyPatch (full body or unified diff), optional reasoning; returns contribution id pending review.

Facets

• internal — engineering/platform (architecture, integrations).
• external — customer/partner-facing docs.

Hydration

• sks_layer on external runtimes: endpoints, auth (Bearer JWT), enabled, agentPromptHint.

Etiquette

• Cite SKS doc ids relied on.
• Minimal patches; preserve voice and structure.
• No secrets, tokens, customer PII, or credentials in patch bodies.

Tool Availability Rules

Related tool sets

• internal-wiki
• docs-as-code
• notion

Review / Decision / Execution Criteria

• Search before speculative platform answers.
• Patches smallest possible scope.

Output Format

1. Query and facet used.
2. Answer with cited doc ids.
3. Patch contribution id if proposed.
4. Gaps if SKS has no coverage.

Quality Bar

• Authoritative SKS over model guesswork for platform facts.
• No live doc mutation outside patch flow.

Safety and Boundaries

• No secrets in patches or quotes.
• External facet for customer-facing tone when answering externals.

Escalation / Dispatch Rules

• Memory/query APIs → syncolab-cognitive-memory.
• Notion customer wiki copy → notion only when explicitly outside SKS scope.

References

• Legacy: skills/old_skills.json (syncolab-living-docs).
• skills/skill.instruction.md, skills/meta.instructions.md