CipherClaw
CipherClaw analyzes recorded agent activity without model calls. It builds causal graphs, scores possible causes, and flags failure patterns.
Project details
- Classification
- Open-source project
- Repository
- Alexi5000/CipherClaw
- Language
- TypeScript
- License
- Apache-2.0
- Status
- Maintained
- Tests
- 3 test notes
- Last verified
- 2026-06-22
- Related write-up
- alexcinovoj.com
What it is
I use CipherClaw to analyze recorded agent activity without asking a model what happened. The same trace produces the same analysis result. It has zero runtime dependencies, and the repository includes runnable examples and tests. Missing trace data limits the analysis.
Architecture
- Trace ingestion builds a causal graph of tool calls, retries, and state transitions.
- Weighted root-cause scoring ranks candidate causes deterministically instead of asking a model to guess.
- Failure prediction flags trace shapes that historically preceded a fault.
- The repository provides an OpenClaw plugin (openclaw.plugin.json) and a library.
Tests and gates
- vitest.config.ts drives the test suite and .github/workflows/ci.yml runs it; see the latest run for current results.
- Zero runtime dependencies, so the dependency surface under test is the code itself.
- Examples directory contains runnable traces used by the suite.
Limits
- Deterministic analysis only reports what the trace records. Missing instrumentation stays missing.
- It explains failures; it does not repair them.
- Scoring weights are tuned for agent traces, not for general distributed-systems tracing.
Source files and examples
Related methods
Acceptance tests for agent work →
Executable checks test the task and detect regressions.
Evaluation tests for releases →
A fixed evaluation suite checks release requirements.
MCP tool permissions →
The server checks which tools a caller may use.
Recording agent runs →
A shared trace ID connects the records of a run.
Spending limits for agent runs →
Budget checks run before provider requests.
Pausing for human approval →
The run waits for approval before the action executes.
Related projects
ClawKeeper →
ClawKeeper checks proposed Claude agent actions against explicit rules before a model call. Autonomous spend defaults to zero, and the database enforces an append-only audit log.
TechTide Harness Kit →
TechTide Harness Kit uses JSON Schema to declare skill permissions, risk tiers, network access, and approval requirements. Automated checks validate declarations and run prompt-injection fixtures.
Swarm 357 →
357 Claude agent roles across six business layers, with portable memory, enforced cost caps, a bash policy gate, and durable checkpoints.
FintheFinder →
FintheFinder pauses research runs for human approval and audits citations deterministically. Its repository includes contracts and test files.
I document the project here. TechTide AI publishes current service scope, prices, and terms.
Production Block Call at TechTide AI→© 2026 Alex Cinovoj · TechTide AI · Columbus, OH