docs: fix drift in Memory Rules & Patterns

[jack-arturo]

First review of best-practices/memory-rules — no prior review-state memory existed for this page. Automated docs-accuracy audit.

Claim	Current state	Fix	Evidence
"API parameters exposed in src/index.ts:764-811"	The six expand parameters are defined at L784–L831; L764–L811 starts before the block and cuts off before expand_min_importance (L818) and expand_min_strength (L825) — two of the six the list documents	Repinned to a permalink covering L784–L831	mcp-automem@2816beb:src/index.ts#L784-L831
{{CURRENT_MONTH}} listed as a common template variable resolving to 2026-02	No such substitution exists anywhere in mcp-automem. The only repo-wide occurrence is a test asserting the cursor template does not contain it	Removed the table row	mcp-automem@2816beb:tests/cli/smoke.test.ts#L565
Tagging schema prescribes a platform tag and a YYYY-MM month tag — in the mermaid diagram, both JSON examples, an "Always include the current month as a tag" instruction, and the Best Practices summary	The shipped memory policy bans both: "no lang/ prefixes, platform tags, or date-stamped tags" and "NO platform tag, NO [YYYY-MM]"; date-scoped facts use t_valid / t_invalid instead	Rewrote the diagram nodes and both examples to bare tags, replaced "Time Tag Conventions" with time_query + validity-window guidance, added a bare-tags caution, fixed the summary bullet	mcp-automem@2816beb:src/memory-policy/shared.ts#L433, #L533, #L876
Importance Scoring Matrix: Preferences 0.6–0.8	Policy stores a stated preference/correction at importance 0.9, which falls outside the documented range. The page also contradicted itself — the Memory Type Taxonomy table above lists Preference at 0.75–0.90	Changed the row to 0.9 and corrected the rationale	mcp-automem@2816beb:src/memory-policy/shared.ts#L543, #L882
Cursor integration pattern: "3-phase lifecycle (conversation start → during work → conversation end)"	automem.mdc.template is built around two-phase session-start recall plus exactly three mid-conversation storage triggers, and its Storage Discipline section forbids storing session summaries — so the "conversation end" phase contradicts the template	Replaced with the two-phase + three-trigger description	mcp-automem@2816beb:templates/cursor/automem.mdc.template#L27, #L96

Verified against: mcp-automem@2816beb7bb81d87bb8d592e8712fd8b4a7cc4f95

Questions

• Namespace tags vs. bare tags. The page's "Namespace Hierarchy" tree (project-x/auth/jwt) and the tag_match: "prefix" example sit awkwardly against the policy line "No namespace prefixes (project/*, lang/*)" — but the same policy file elsewhere recommends "precise, namespace-style tags (e.g. slack/channel-ops)". The source is genuinely ambiguous here, so I left both sections alone. Worth settling upstream in shared.ts and then reflecting here.
• Hook system history. The note "removed in v0.8.0 and v0.6.0 respectively" could not be checked against the current tree (the hook system exists again today, in a different, LLM-judged form). Left as written.

Unverified

• "This philosophy evolved from early experiments with hook-based automation… the instruction-based approach proved more reliable" — a narrative claim with no code anchor. Left alone.
• Claude Desktop "Personal Preferences (Settings → Profile → Personal Preferences)" — a third-party UI path, not verifiable from these repos.

Follow-ups

Beyond the 5-fix cap:

• "Template File Locations" tree is stale. It shows 3 markdown files plus cursor/ and codex/. templates/ now also contains antigravity/, claude-code/, copilot/, hermes/, openclaw/, COPILOT_INSTRUCTIONS_MEMORY_RULES.md, claude_desktop_config.json, and cursor_mcp.json; cursor/ holds three files (automem.mdc.template, rule-evals.md, user-rules.md), not one. Nothing listed is wrong, so this is a completeness gap rather than an error.
• tag_match: "prefix" framed as opt-in. "Set tag_match: \"prefix\" to support hierarchical tag queries" reads as though prefix matching must be enabled; the tool schema documents prefix as the default. Minor wording.
• Memory Type Taxonomy importance ranges are looser than the policy's point values (Decision 0.85–0.95 vs 0.85-0.9; Pattern 0.70–0.85 vs 0.8; Insight 0.75–0.85 vs bug-fix 0.75). Each documented range contains the policy value, so none is provably wrong — but the two tables on this page could be reconciled to the policy's concrete numbers.

---

Generated by Claude Code

[cloudflare-workers-and-pages]

Deploying automem-website with    Cloudflare Pages

Latest commit:	55b9cf0
Status:	✅  Deploy successful!
Preview URL:	https://fe7f674b.automem-website.pages.dev
Branch Preview URL:	https://docs-audit-best-practices-me-fi45.automem-website.pages.dev

View logs