Skip to content

Agentic Dev 0.15.0

Latest

Choose a tag to compare

@szaher szaher released this 02 Oct 19:50
a4c25e4

v0.15 (Agent Ready remediation + CI): slices 1–4.

The readiness exit codes below are frozen as part of the automation contract from this release.

  • The generated readiness.ci-check workflow uses actions/checkout@v6 (with persist-credentials: false) and actions/setup-python@v6, matching this repository's own CI.

  • Added agentic ready make [PATH] --target LEVEL [--dry-run] [--ci-check] [--max-rounds N] [--json]. It runs bounded rounds of plan → safe remediation → reassess and stops on human decisions (listed with candidates) instead of guessing. --dry-run previews the whole run in a throwaway copy. Exit codes: 0 target met, 1 conflict, 2 usage error, 3 rolled back, 4 stopped. New contract: readiness-make-v1.

  • Added agentic ready verify --target LEVEL [--spec-version V] [--spec-sha256 H] [--scope ci|local] [--json], a CI gate with stable exit codes: 0 met, 1 not met (regression), 2 usage or spec error, 3 pinned spec mismatch. It emits the readiness-verification-v1 contract, which names the blocking rules and reports the local vs CI-visible difference with the untracked files responsible.

  • Added the assessment scope (local / ci). In ci scope only tracked files are evidence, so untracked or Git-excluded local state (for example a locally generated AGENTS.md) never counts. Every readiness document reports its scope. assess, explain, and plan accept --scope.

  • Added the opt-in readiness.ci-check maintenance action (ready apply --ci-check). It writes a managed GitHub workflow that verifies today's CI-visible level, pinned to the exact Agentic Dev version and spec. It is evidence-neutral, read-only, and credential-free.

  • Added the remediation contract (docs/REMEDIATION.md) and schemas/readiness-remediation-v1.schema.json. Rule remediations (safe-automatic / human-decision / unsupported) are capped by the Agent Ready spec classification and never upgraded. Separate maintenance_actions provide operational scaffolding that must be evidence-neutral, never reference rules, use only allowlisted targets, and never add credentials, permissions, or installers. changes are managed blocks with normative apply, conflict, idempotency, and rollback semantics.

  • Added agentic_dev.readiness.remediation.propose, which builds a read-only remediation preview with change provenance and detected candidates (never selections) for policy decisions.

  • Documented the local vs CI assessment scope requirement for ready verify.

  • Added agentic ready diff and agentic ready apply [--dry-run] (v0.15 slice 2). They write only safe-automatic changes as managed blocks, all-or-nothing, with per-file snapshots, atomic writes, and full rollback. Runs are idempotent, preserve bytes outside blocks, refuse hand-edited blocks, symlinks, and CRLF files, refresh stale managed blocks, and flag untracked targets. Exit codes: 0 ok, 1 conflict (nothing written), 2 usage error, 3 rolled back.

PyPI: uv tool install agentic-dev==0.15.0