AI-assisted project development and tracking: a typed, versioned design record for engineering projects, driven by your agent over MCP.
Redraft demonstrates solid definition quality with 24 well-named, verb-prefixed tools and comprehensive parameter schemas. All tools have descriptions (avg ~150 chars, within baseline 34-392 range). Input schemas are present and typed for all tools. However, output schemas are not documented in the provided source, and several tools lack guidance on error recovery or prerequisites. Tool composition is strong (each does one thing), but descriptions could be more LLM-optimized with explicit 'when to use' guidance. No security issues detected (no secrets in params). Error handling via tool_errors.py is well-structured but not surfaced in tool descriptions.
Collapse root_id's part_of subtree into a report bundle: nested sections with attached cross-links, decision_tables grouped by driver question/requirement (any status, cross-branch — see report-bundle-v2.md), and per-section gaps (open questions, decisions lacking rationale). The agent writes the narrative; this never summarizes.
One-call topic guidance: hybrid search hits, their 1-hop neighborhood, open questions, and unjustified decisions scoped to the topic. Prefer this over grepping graph files for "what do we know about X" -- it returns the matched nodes together with their neighborhood and open gaps, which grep cannot assemble.
Groups of >=2 node ids identical under Unicode NFC+casefold.
{a, b} node pairs linked by a `contradicts` edge (a dangling contradicts edge is surfaced by dangling_edges() instead, not here).
Create an edge from src to dst with the given type.
Create a new node. Does not attach it anywhere -- call create_edge next.
Output schemas not documented in tool definitions. LLMs cannot plan downstream calls or extract required fields without knowing response structure.
Destructive tools (delete_node, merge_nodes, rename_node, snapshot) lack confirmation/dry-run guidance in descriptions. No mention of irreversibility or recovery steps.
Query tools (search_nodes, find_similar, briefing) lack pagination parameters (limit/offset/cursor). Large result sets could exhaust context window.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 64 | 2026-07-28+ | v2 |
Edges whose src and/or dst no longer resolves to a node.
Decisions (any status) with no inbound `justifies` edge.
Delete an edge from src to dst of the given type.
Delete a node and all its edges. Inbound wikilinks in other nodes' bodies are rewritten to remove the deleted node's id.
Vector-only near-duplicate search -- the dedup primitive. text_or_id may be an existing node's id (reuses its cached vector, excludes it from its own results) or free text (embedded fresh). Prefer this over grepping graph files to check "is this already recorded" -- it matches paraphrase across different wording, which grep cannot.
Fetch a node. neighbor_depth > 0 also returns nodes reachable within that many hops (any edge type, any direction) and the node's own direct edges.
Fetch all nodes reachable from root within max_depth hops and all edges among them, optionally filtered by edge type and direction.
Merge keep_id and drop_id: drop_id's edges are relinked to keep_id, drop_id is deleted, and keep_id's body is optionally appended with drop_id's body.
Fetch nodes reachable from a root within max_depth hops, optionally filtered by edge type and direction.
Question nodes with status='open'.
Nodes with zero edges in any direction/type (a legitimate root-level concept with no part_of parent is NOT flagged for that alone).
Cheap, shallow map of the project's shape: every spine root, each root's direct part_of children as branches with per-subtree tallies (descendants, open questions, decisions by status, unjustified decisions), whole-graph totals, and the top open questions. Prefer this over grepping graph files for the project's shape -- it returns the spine structure and tallies in one cheap call, which grep cannot compute. Call this (or read graph://project/overview) at the start of a session, before anything else, to load the project's shape and make your first real call targeted instead of blind.
Rebuild the derived index from graph/nodes/*.md. Run after a manual git pull/merge, or to recover from a deleted/corrupted index/ dir.
Rename a node's id (title stays unchanged). Inbound wikilinks in other nodes' bodies are rewritten to use the new id.
Hybrid FTS+vector search, fused by Reciprocal Rank Fusion. types/status filter each branch's candidates before fusion (a filtered-out result never consumes a fusion rank slot it wasn't really competing within). Prefer this over grepping graph files for ANY lookup -- it matches paraphrase, filters by type/status, and returns typed nodes, which grep cannot.
git add (pathspec-scoped) + commit; push only if push=True (defaults off to avoid a surprise network operation).
Nodes untouched since before_iso. types/statuses omitted (None) use the default non-terminal question/decision/milestone set; an explicit [] matches nothing (not reinterpreted as 'omitted').
Update a node's title, body, status, or properties. body_mode='replace' (default) overwrites; 'append' concatenates.
Retrieval tools (decisions_without_rationale, open_questions, orphans, contradictions, dangling_edges, case_collisions) have minimal descriptions (<80 chars). LLMs lack context for when to invoke them.
Error handling is implemented in tool_errors.py but not surfaced in tool descriptions. LLMs cannot know which errors are retryable or what recovery steps exist.