fxspeiser/crosscheck-agent

tools

• actionability

  low

• audit

  Post-run rubric scoring. Audits an output (from output_to_audit, or pulled from the latest transcript for session_id) against a rubric. The auditor is selected to exclude the producing_panelists so a

• bench

  Run repo-scoped goldens (*.json in .crosscheck/goldens/) against a panel; rule-based verifiers (contains / regex_match / contains_any / contains_all / not_contains / min_length) score each provider, a

• confer

  Ask one or more providers the same question in parallel; return every answer.

• constraint_adherence

  high

• coordinate

  Structured Proposer → Critic(s) → Synthesizer flow. Each role emits a JSON envelope; the synthesizer emits a validated StructuredSynthesis. Persists key claims to the SQLite claim-list with supports/a

• covers_open_questions

  med

• critique

  Have each panelist list the top weaknesses of a proposed answer / plan / code with severity tags (low

• debate

  Bounded round-trip debate; the configured moderator synthesises a result. Optional structured: true makes the synthesis a JSON-schema-validated object (consensus, dissent, key claims, citations, open

• delegate

  Cross-model handshake: route a confer or review call through one named provider, with quota tracking by session_id and by requesting provider. Refused calls return accepted: false with a structured re

• explain

  Replay a session as a navigable tree with cost / latency / token annotations per tool call and per provider call. Reads usage_log and persisted transcripts. Returns both a structured envelope and a pr

• factual_grounding

  high

• fetch

  Retrieve a URL with deny-by-default allowlist (fetch.url_allowlist prefix list) and persist a sha256-keyed snapshot under .crosscheck/evidence/. Cached on repeat unless force_refresh: true. Use to gro

• id

  severity

• internally_consistent

  med

• list_providers

  Discover which LLMs are currently available and which are in the active set.

• max_dag_nodes

  / max_dag_depth — protect orchestrate from a planner that emits an unboundedly wide or deep DAG.

• max_session_cost_usd

  cumulative dollar cap.

• max_session_tokens

  cumulative token cap.

• max_session_wall_seconds

  wall-clock cap.

• no_pii_leak

  high

• orchestrate

  Plan-then-execute across sub-agents. The moderator decomposes a goal into a DAG of subtasks (or you pass a pre-authored dag), workers run in parallel where deps allow, and a final synth pass recombine

• pick

  Multi-criteria decision-making. Each provider scores every option on every criterion (0..1); the tool aggregates with criterion weights, returns a ranked list, and surfaces the top-K cross-provider di

• plan

  Collaborative step-by-step planning with risks + alternatives. Honours structured: true.

• recall

  Full-text search across persisted transcripts via SQLite FTS5. Pull prior context out of past sessions without re-running expensive panel calls. Ranks by bm25, returns windowed snippets with [[ ]] hit

• recommend_panel

  Smart router as a tool. Returns a recommended provider lineup for a given purpose based on accumulated usage_log reliability + cost + engagement, with optional prefer_provider / avoid_provider hints.

• review

  Peer code / proposal review across one or more LLMs. Pass untrusted_input: true when the snippet may contain prompt injections.

• scoreboard

  Read-only snapshot: per-provider weight + wins/losses/abstains + delegations, plus totals for sessions/claims/links/delegations and (optional) the last N redacted event lines. The data the UI panel re

• solve

  Iterative propose → verify → retry. Provider drafts a literal solution; a shell (sandboxed subprocess: timeout, RLIMIT_AS, RLIMIT_CPU, isolated tmpdir) or regex_response verifier accepts or rejects; f

• triangulate

  Run a coordinate flow and return consensus + minority report plus per-provider weights drawn from accumulated bench / critic-ballot win-rate. The "give me one answer, but be honest about disagreement"

• update_crosscheck

  Compares your local git HEAD against main at https://github.com/fxspeiser/crosscheck-agent. With apply: true, runs git pull --ff-only in the install directory; the server can't reload itself, so the r

• verify

  Run a list of deterministic property checks against caller-supplied data — no LLM calls. Supports text-pattern kinds (contains, not_contains, regex_match, contains_any, contains_all, min_length), shel