Developer Tools for Auditable AI Agents
Use EDGAR MCP access, deterministic CLIs, webhooks, PBIX lineage, and machine-readable data when agent outputs must trace back to source.Short answer
CMD+RVL gives developers source-aware tools for agents that query SEC filings, receive outcome changes, inspect PBIX lineage, and cite machine-readable data.Start with the integration surface
These tools are for workflows where an agent, script, or internal app must show which source records supported an output.Query SEC filings from AI agents
Use the private-beta EDGAR MCP server with Claude, Codex, and MCP-compatible clients to ask source-aware questions about filings, companies, form types, and dates.Request EDGAR MCP accessReceive outcome changes by webhook
Send state-transition notifications into a local service or internal workflow, then inspect the JSON payload and attached evidence record.Read the webhook quickstartExtract Power BI model lineage
Turn PBIX model structure, DAX expressions, M code, and lineage into an LLM-ready metadata export for review and handoff.Try the PBIX lineage extractorGive agents machine-readable context
Use structured-finance facts, outcome definitions, and glossary terms in JSON form so agent outputs can cite stable, source-ready context.View machine-readable dataQuick start: what changed between two quarterly reports?
# What's in each quarterly report?
vacuum q3/ q4/ | hash | lock save --name quarterly
# What actually changed?
rvl q3/positions.csv q4/positions.csv
# -> 3 of 847 rows changed, net +$2.1M notional, 1 new position
# Seal the evidence
pack create --lock quarterly.lock --include q3/ q4/
# -> pack-a7f3b2.pack (content-addressed, tamper-evident)
How the tools work
Every CMD+RVL tool follows four rules:1
Structured outputEmits JSON with a version field, a domain result, and a structured refusal when the tool cannot operate.
2
Self-description--describe emits a machine-readable operator manifest. --schema emits the output JSON Schema.
3
WitnessAppends a content-addressed, hash-chained record on every invocation. Every run is auditable.
4
DeterminismSame inputs, same output, every time. No randomness, no side effects, no network calls in the verification path.
Protocol Tools
Composable Rust CLIs. Each tool does one thing. Agents decide what to run. These tools decide what the evidence supports.vacuum (what's there?) -> hash (prove identity) -> fingerprint (recognize templates) -> lock (pin inputs) -> shape (comparable?) -> rvl (what changed?) -> pack (seal the evidence)
Open Source
vacuum
Enumerates artifacts in scope. Emits a deterministic sorted JSONL manifest with size, mtime, and MIME type.View on GitHubOpen Source
hash
Streaming content hashing. Adds SHA-256 or BLAKE3 byte identity to every artifact in a manifest.View on GitHubOpen Source
fingerprint
Template recognition. Tests artifacts against versioned assertion-based definitions and produces content hashes.View on GitHubOpen Source
profile
Column-scoping configs for report tools. Draft/freeze lifecycle, deterministic key suggestion, schema linting.View on GitHubOpen Source
lock
Dataset lockfiles for data. Pin artifacts, fingerprints, and tool versions into a self-hashed, tamper-evident, reproducible snapshot. Includes lock verify for integrity checks.View on GitHubOpen Source
shape
Structural comparability gate. Know whether two CSV datasets can be compared before you waste time trying. Checks schema overlap, key viability, row granularity, and type consistency.View on GitHubOpen Source
rvl
Reveals the smallest set of numeric changes that explain what actually changed between two CSV files. Ranked explanations, tolerance-aware, three clear outcomes: REAL CHANGE, NO REAL CHANGE, or REFUSAL.View on GitHubOpen Source
canon
Deterministic entity resolution. Resolves identifiers against versioned registries with a reviewable source trail.View on GitHubOpen Source
pack
Evidence sealing for lockfiles, reports, and tool outputs in one immutable, content-addressed evidence pack. Self-hashed manifest with pack verify for integrity checks.View on GitHubIn Development
Coming Soon
verify
Invariant checks for single-artifact rules (JSON) and cross-artifact constraints (SQL via DuckDB).Coming Soon
benchmark
Extraction accuracy scoring. Checks (entity, field, value) assertions against candidate datasets.Coming Soon
compare
Exhaustive cell-by-cell diff without materiality compression.Coming Soon
assess
Decision framing for PROCEED / ESCALATE / BLOCK against declared policy.SEC EDGAR & Financial Data
Open Source
EDGAR Change Interpreter
A Claude skill for identifying material changes, silent risks, and interpretation traps in SEC filings (10-K, 10-Q, 8-K).View on GitHubPrivate Beta
EDGAR MCP Server
Private-beta SEC EDGAR MCP server for Claude, Codex, and MCP-compatible clients that need source-aware filing metadata.Request EDGAR MCP accessLive Agent
Fee Exhibit Fragility Authority
Live Kovrex agent for SEC/EDGAR filing-fee exhibits. It checks accepted filings for deterministic reclassification-risk conditions and returns bounded findings or structured refusals; it is not a general filing validator.View on KovrexInstall all protocol tools
brew install cmdrvl/tap/vacuum
brew install cmdrvl/tap/hash
brew install cmdrvl/tap/fingerprint
brew install cmdrvl/tap/profile
brew install cmdrvl/tap/lock
brew install cmdrvl/tap/shape
brew install cmdrvl/tap/rvl
brew install cmdrvl/tap/canon
brew install cmdrvl/tap/pack