agent-harness GitHub
Menu

Overview

agent-harness Agent Harness Blueprint Blueprint plugin product brief The Harness Book Deny-floor known limitations Licensing scope and decision Replay tool product brief Agent Harness roadmap Harness Specs

Docs

Advisory agent-operations measurements Mapped authority-document drift probe Mapped authority drift: pinned local trial Benchmarks and measured evidence Corpus path boundary Advisory entropy measurements Learning evaluator evidence boundaries Outcome stream validation System comparisons and canary evidence Advisory logical-job accounting Replay lab Python check inventory Replay CLI input identity and timeout limits Replay input decoding failures Replay report output boundary Replay process policy-root aliases Replay supervisor marker scope Review evidence and acceptance oracles Offline reviewer-packet diagnostics System state Process-fixture startup is not containment evidence

Agentic

Architecture additives — perfect assistant → estate owners Perfect autonomous agentic assistant — estate synthesis

Evals

Harness extension notes Evaluation merge-boundary contract: v0 Merge-gate model Outcome ledger Review under flood Agentic development evaluation taxonomy: v0

Evals / Examples

Worked evidence packet: a private product PR

Extraction

Replay v0 extraction inventory

Freeze

Legacy deny-floor v1 freeze record

Maintenance

Managed Codex agent publication Sync input and Doctor session roots

Observability

Coding-agent loop observability v0 Offline source-record identity inspector

Observability / Adapters

Claude Code to agent-loop v0: mapping note

Review evidence

Review/oracle wave: source ledger

UX evaluation

Frozen observation verifier Evidence-bound issue seeds Advisory product UX evaluation Offline scenario binder Scenario authoring contract v0 Taskdeck-first adapter sketch Workspace handoff acceptance profile Workspace receipt lab

Replay CLI input identity and timeout limits

Follow-through for #146 and the CLI timeout portion of #145. Decision-output JSON handling is separate in #334; corpus path containment landed in #333.

Process labels

The CLI emits process-<full identity SHA-256> as each process-policy ID instead of copying the filename stem. A readable policy with a space, leading underscore, leading dot, non-ASCII letter or long stem can therefore satisfy the existing portable manifest-ID grammar without renaming the file.

This reuses the existing v9 identity digest. The supplied filename, bytes, parent tree, executable, timeout and other bound inputs still contribute to that digest. Distinct names are not collapsed by lossy sanitization. This is not a location-independent content ID or authentication of an execution.

Compatibility: Process-source labels change, including previously accepted filename stems. For identical semantic inputs, the existing identity SHA-256 and derived run ID remain stable: derive_run_id() uses the source digests, not either source label. Manifest bytes change because the label is recorded; that is not a new semantic run. The manifest schema, recorded-source IDs and validation of existing reports are unchanged. Correlate archived and new runs using their run IDs and declared semantic inputs, not the former stem label. Commas and multiline argv values remain outside the existing grammar (#142).

CLI wait input

--timeout accepts finite seconds in 0 < timeout <= 86400. The existing 30-second default and fractional positive values remain supported. One day is an explicit CLI upper bound, not a benchmark or a new cleanup timeout.

Invalid values, including an extreme finite value such as 1e100, fail argument parsing with exit 2 before reading a corpus, loading a policy or publishing reports. The diagnostic states the accepted bound without a Python traceback. The process runner, constructor API, snapshot behavior and cleanup grace are unchanged; this does not claim to normalize every platform-wrapper exception. The remaining #145 cases stay open.

Verification

python -m unittest replay_v0.tests.contract.test_cli_inputs -v exercises real filename/manifest round trips, repeated fixed-time runs, distinct names and semantic changes. Timeout controls prove early rejection and include a real CLI subprocess whose candidate launch marker and report directory remain absent. The module is included in the existing three-OS source-check lists and in replay discovery. No new dependency, job, privilege or model-based gate is added.