Index rules from a markdown file directly into SQLite.

mode="atomic" upserts canonical blocks by code (existing rules are updated in place, new ones created). mode="legacy" stages each old-style block as a pending_proposals entry instead of writing to rules directly.

Access level: write. Returns: JSON with indexed/created/updated counts, a per-scope breakdown, and any parser warnings or errors.

Index lessons from a markdown file directly into SQLite.

Same logic as index_rules_from_markdown but targets the lessons table.

Access level: write. Returns: JSON with indexed/created/updated counts, a per-scope breakdown, and any parser warnings or errors.

Convert a legacy markdown file into atomic-format pending_proposals, one per legacy block, with best-effort scope inference from each block's text. Never writes .md files directly — always stages for human review via approve_proposal.

Access level: write. Returns: JSON with the number of proposals created and their id/type/ scope_id/suggested_file.

List pending proposals for human review, optionally filtered by project, type, or status.

Access level: read. Returns: JSON array of matching pending_proposals rows.

Approve a pending proposal. For type='rule'/'lesson' this assigns a new RN-/LL- code, appends the atomic block to the scope's .md file, and inserts the corresponding row; for type='update' it splices the target's existing block in place instead. This is the single choke point that writes proposal content into both the .md file and SQLite together.

Access level: write. Returns: JSON with the resulting code, scope_id, and file location (file_path/file_offset/byte_length).

Update the proposed_text and/or metadata of a still-pending proposal; its status stays 'pending'.

Access level: write. Returns: JSON with the proposal_id and its status.

Reject a pending proposal. No side effects besides updating the record.

Access level: write. Returns: JSON with the proposal_id and its status.

Promote a rule to a more general scope: marks the original as deprecated (in both SQLite and its source .md file) and records a PROMOTED rule_history entry. Does not create the rule in the new scope — that requires a separate proposal.

Access level: write. Returns: JSON with the rule_id, new_scope_id, and resulting status.

Create a proposal to deprecate a rule or lesson on its own, without changing its scope (unlike promote_rule). Accepts both RN-* and LL-* ids. Only creates a pending_proposals row (type='deprecate') — no immediate effect; approve_proposal performs the actual deprecation after human review.

Access level: analyze. Returns: the new proposal's id, as a plain string.

Create a new project scope and its knowledge base directories: a 'project-' scope under parent_scope, plus knowledge-base/projects/ and lessons/projects/ directories with a placeholder rules .md file.

Access level: write. Returns: JSON with the new scope_id, name, parent_scope, and created file_path.

Generate or update ChromaDB embeddings for active rules/lessons whose embedding_id is still NULL — an incremental backfill, not a full re-embed.

Access level: write. Returns: JSON with processed/skipped/errors counts and duration_seconds, or {"error": "RAG_NOT_INSTALLED", ...} when the rag extra is absent.

Generate a Claude Code skill for a project from its resolved rules: writes .claude/skills/meridian/<project_id>/SKILL.md and references/rules.md under project_path, and updates its AGENTS.md.

Access level: write. Returns: JSON mapping "skill_md"/"rules_md"/"agents_md" to their written file paths.

Query active rules visible to a project across its full scope hierarchy (most-specific scope first), via plain SQL filtering or — when query_text is set and embeddings exist — semantic search.

Access level: read. Returns: a JSON or TOON array of matching rules.

Query active lessons visible to a project across its full scope hierarchy (most-specific scope first), via plain SQL filtering or — when query_text is set and embeddings exist — semantic search.

Access level: read. Returns: a JSON or TOON array of matching lessons.

Return a rule's full text, its complete rule_history, and any lessons linked to it.

Access level: read. Returns: JSON with rule, history, and linked_lessons.

Return a rule's chronological history and linked-lesson summary without loading full text — lighter than get_rule_context.

Access level: read. Returns: JSON with rule_code, scope, severity, category, history_events, and linked_lessons_summary.

Resolve and cache a project's effective scope chain (most-specific first) with each scope's dynamic attributes (ADR-002).

Access level: read. Returns: JSON with project_id and the resolved scopes list (each with its attributes).

Report this running server process's version/commit, and whether the code on disk has changed since this process started — Python doesn't hot-reload, so an editable install that was git pulled without restarting the server keeps serving the old code with no other visible sign of it (Hallazgo 7).

Access level: read. Returns: JSON with version, the commit captured at server startup, a freshly-read commit as of this call, and a stale flag (true only when both commits are known and differ) with a restart note.

Return the chronology of rule_history changes for a project or scope, plus any rules currently deprecated in that scope.

Access level: read. Returns: JSON with events and deprecated_rules.

Audit a PR diff against a project's active rules and lessons (detail="full"): persists an audit record (SHA-256 diff hash + a snapshot of the rule versions evaluated) and returns the diff plus rules/lessons for the calling LLM to classify.

Access level: analyze. Returns: JSON with audit_id, pr_diff, rules, lessons, and an instruction_for_client.

Link PR review feedback to the most recent audit_pr record for the same pr_ref/project_id, returning the rules that applied at that time for the calling LLM to classify each feedback point.

Access level: analyze. Returns: JSON with feedback_text, rules_that_applied, pr_audit_record, and an instruction_for_client.

Check a planned feature description against a project's active rules and lessons (detail="full"); persists a planning_checks record and returns the context for the calling LLM to flag conflicts.

Access level: analyze. Returns: JSON with feature_description, rules, lessons, check_id, and an instruction_for_client.

Return everything the calling LLM needs to extract candidate rules from a transcript: the clean text, the project's scope chain with attributes, existing rules for duplicate detection, and the canonical rule template. Does not itself create any proposal — call create_pending_proposal per candidate.

Access level: analyze. Returns: JSON with clean_text, scopes, existing_rules, template, and an instruction_for_client.

Return everything the calling LLM needs to extract candidate lessons from a transcript: the clean text, the project's scope chain with attributes, existing lessons for duplicate detection, and the canonical lesson template. Does not itself create any proposal — call create_pending_proposal per candidate.

Access level: analyze. Returns: JSON with clean_text, scopes, existing_lessons, template, and an instruction_for_client.

Validate and persist a pending proposal of type 'rule', 'lesson', or 'update' (private tags stripped from proposed_text and metadata). The body goes in proposed_text and the other block fields in metadata; a full rendered atomic block in proposed_text is split into the two. For type='update', target_id must reference an existing RN-/LL- id and the proposal inherits that target's scope_id. Does not write to rules/lessons or any .md file — approve_proposal does that after human review.

Access level: analyze. Returns: the new proposal's id, as a plain string.

Return the canonical ## RN-XXX-NNN atomic block template, its field descriptions, and usage instructions.

Access level: read. Returns: JSON with template, fields, usage, and (if project_id given) scope_suggestions.

Return the canonical ## LL-XXX-NNN atomic block template, its field descriptions, and usage instructions.

Access level: read. Returns: JSON with template, fields, usage, and (if project_id given) scope_suggestions.

Return the combined rule+lesson template guide and workflow instructions for turning a transcript into pending proposals.

Access level: read. Returns: JSON with template, fields, usage, and (if project_id given) scope_suggestions.