v0.3.4

0.3.4 (2026-11-10)

Features

• dev: add adr-lint to make check (58c5307)
• dev: adr-lint checks the engine index and mentions of missing ADRs (2a492e5)
• dev: adr-lint checks the engine index and mentions of missing ADRs (090de46)
• engine: DOC-04 flags superseded, deprecated or rejected ADRs (568bc5c)
• implement: add a prose check template for target repositories (b4e5fbe)
• implement: bring the writing rules to target repositories (v0.3.6) (8ec7fee)
• record the ADR policy in ADR 0009 and check it (DOC-04, adr-lint) (c304ee5)

Bug fixes

• engine: add the tool-07-guard-no-exit eval file to the secrets baseline (54c10e0)
• engine: EXEC-01 reads every module of a pinned ecosystem (86cab41)
• engine: TOOL-07 needs a guard that exits non-zero (71b4e61)
• eval: regenerate the eval-expected files with the current paths (0ccd527)
• implement: store commit ids in the state as 12 digits (ddf8b05)
• skills: resolve every link the audit bundle keeps (6a98a5d)

Documentation

• add the agent-tuned ADR policy and a lean template (46e6577)
• an agent-tuned ADR policy, a lean template and every ADR converted (b735040)
• convert the audit ADR to the lean template (274b16d)
• convert the engine ADRs to the lean template (a1327e5)
• convert the implement ADRs to the lean template (81ab990)
• convert the root ADRs to the lean template (1d3b048)
• convert the spec decision records to the lean template (221fff3)
• describe how the implement recipe brings the writing rules to a target (907fd76)
• give ADR 0005 the current reason to reject a back-merge from main (111fedb)
• greenfield: ship the ADR policy and the lean template (83ce5b8)
• implement: add plain-language and diagram rules to the style guide (4d78aea)
• implement: add the Dependabot pre-commit ecosystem to step 8 (9db8c0b)
• implement: confirm the GitLab variables of the source-branch job (18ec3b3)
• implement: count only warnings in the target's prose ratchet (9c7ba80)
• implement: limit prose check copies to three permissive licences (d2bf4ca)
• implement: name what triggers the AGT-12 and DOC-05 steps (31f9a9a)
• implement: record existing prose debt instead of rewriting it (4a1f495)
• implement: use the shared message layouts in the two playbooks (2aa044e)
• implement: write a "How to write for people" section in adapt-agents-md (87940ae)
• implement: write ADRs into targets under the ADR policy (8abd227)
• implement: write the Renovate security-only form from Renovate's docs (af0b6a1)
• license the greenfield templates under MIT-0 (786fba5)
• license the greenfield templates under MIT-0 (93c6b66)
• point every agent file and router at the ADR policy (ad0e4b5)
• record why ADRs hold only the current decision (ADR 0009) (30ac7e5)
• replace ADR index status and date columns with summaries (9bf7534)
• restore the two style guide headings an edit merged (c9dc09f)
• spec: let any verb run the dependency scan in REQ-SH-030 (5182077)
• word the AGENTS.md ADR pointer as a rule, not a deferral (52fe131)

CI and release

• run the eval-expected check in make check (d25eae4)
• shard the dev checks in the Python 3.14 test parts (2dd517d)
• shard the dev checks in the Python 3.14 test parts (4bac0ca)

v0.3.3

0.3.3 (2026-11-10)

Features

• audit: open the rendered report with a bottom line (52768e2)
• audit: record a check newer than the engine's ruleset as UNKNOWN (4c9c82d)
• audit: record the framework release and install in each report (61d0e2d)
• dev: agent tools for this repository: one .mcp.json, a first-run offer, language servers (109e7ad)
• dev: assert the pull request body layout in the implement eval (51bddf9)
• dev: check that each recipe SKILL.md links the user prompt (1a569d0)
• dev: document language servers beside the code index (07d42bb)
• dev: give prose-check issues a severity and always exit 0 (d36a954)
• dev: offer each contributor a one-time agent tool setup (6dce4b1)
• dev: register codebase-memory-mcp in one project .mcp.json (683ddb3)
• dev: translate .mcp.json into each harness's MCP config (b0b2085)
• engine: add AGT-12 and DOC-05, the v0.3.4 Track T checks (1703aa2)
• engine: add AGT-12, communication rules stated (5d16098)
• engine: add CI-05, workflow files are linted (a86ddab)
• engine: add DOC-05, docs are readable (fc80971)
• engine: add EXEC-08, tool versions visible to dependency automation (7172596)
• engine: add TOOL-07, a local guard against commits to a protected branch (17bd680)
• engine: add TOOL-07, a local guard against commits to a protected branch (v0.3.3 C5.2) (2fc92ec)
• engine: AGT-09 names every branch the declared rules protect (910b718)
• engine: AGT-09 names every branch the declared rules protect (1dc8bab)
• engine: HYG-09 reads ecosystem coverage and the target-branch trap (c035331)
• engine: HYG-11 reads ruleset files and the source-branch check (23f7146)
• engine: HYG-11 reads ruleset files and the source-branch check (v0.3.3 C5.1) (4b93763)
• engine: open the Markdown report with a bottom line (6afaba3)
• engine: TOOL-04 reads CI as well as the local verbs (36aa387)
• engine: TOOL-04 reads CI as well as the local verbs (9dfc3ae)
• engine: TOOL-06 also reads the commit rule in CI (d7ad429)
• fix the layout of the recipes' questions and reports (3993263)
• give both recipes one prompt on how to talk to the user (cf7a101)
• greenfield: check commit messages in CI as well as in the hook (9f693fe)
• implement: record in the state which framework created and last wrote it (23b354c)
• implement: stop on a report from another schema or a newer release (2309763)
• plain communication in the recipes and a report bottom line (v0.3.4 R, H) (b990687)
• stack-adaptive hook guidance and CI-05 workflow lint (v0.3.3 Track C: C0, C1) (66479e7)
• stamp the framework release in reports and states, and refuse files the tools cannot read (2047b4f)
• stamp the release version in reports, state and skill bundles (5d4ff90)
• TOOL-06 reads the commit rule in CI as well as in the hook (v0.3.3 C4) (5232740)
• v0.3.3 Track C2 and C3 (HYG-09 reads the update policy, new check EXEC-08) (a701a04)

Bug fixes

• audit: list what the scan could not settle as sub-bullets (80e7bbc)
• dev: ignore .env files at any depth (f807d63)
• dev: keep each osv-scanner ignore file next to the lockfiles it covers (8b64cc1)
• dev: name more harnesses and move the first-run trigger to the top (22ba4a6)
• engine: EXEC-01 and TOOL-04 read nested pins and Makefile chains (4f1f1bf)
• engine: EXEC-01 reads the toolchain pin in a nested manifest (f2ff3bd)
• engine: EXEC-07 counts types per language and reads a pyright mirror (26fd6f5)
• engine: EXEC-07 counts types per language and reads a pyright mirror (73fa46b)
• engine: TOOL-04 follows Makefile variables and prerequisites (2e00e34)

Documentation

• add a diagram to each explanation page and tutorial (d965909)
• add a purpose line and a diagram to the page templates (9506b0c)
• add generated audit and implement master lists (b3a9377)
• add generated audit and implement master lists (01be98d)
• audit: an opt-in live read of the branch rules for...

v0.3.2

0.3.2 (2026-11-09)

Features

• Improved communication guidelines for genAI (aa03ab8)

Documentation

• add communication rules and a plain-language prose check (91bc77b)
• move COMMUNICATION_RULES.md into docs/ (7acfcab)
• record automatic deletion of merged branches (9034eb4)
• record automatic deletion of merged branches (058c710)
• rename COMMUNICATIONS.md to COMMUNICATION_RULES.md (ccf8c5c)
• shorten the long sentences in the release process pages (8195c59)

v0.3.1

0.3.1 (2026-11-08)

Bug fixes

• engine: report the released version 0.3.0 instead of 0.1.0 (1214f59)

Documentation

• add a development tooling reference page (7e19468)
• add a step-by-step guide to contribute a change through dev (5806e9e)
• add the harness disclaimer and the report steps to the README quick start (7a8f4eb)
• clone main in the README quick start (f1d41a2)
• describe the dev and main branches in the README (bdcbe3d)
• describe workflow-lint, commit-lint and CODEOWNERS (2aa6632)
• fix the overview link and the skill list in the README (00d4ffc)
• make dev the default branch in ADR 0002 and the process docs (1a557dd)
• recommend a fresh session before the implement step in the README (0efe17c)
• record the Dependabot policy and the dev tool pins (ADR 0003) (df049b5)
• record the dev and main branch model and the promotion release (268ed75)
• record the dev ruleset as active (e417e1a)
• release: describe the release-please process (ae0f2ed)
• reword the README introduction (0fdcd4a)

Build and dev tools

• dev: move the dev tool pins into manifests that Dependabot reads (6002b89)

CI and release

• dependabot: open security pull requests only, never for a major version (0cfcc6c)
• lint the workflows and the commit subjects in the gate (e83a10f)
• release: automate versions and releases with release-please (be5d86f)
• release: publish only a tag whose commit is on main (195d76b)
• require pull requests into main to come from dev (8c3c05a)
• run CI on pushes to dev and target Dependabot at dev (58d65c1)

v0.3.0

What you can do with v0.3.0

Make an existing repository safe for AI coding agents, with two skills for Claude Code.

1. Clone this repository and start Claude Code in it.
2. Say: "Audit /path/to/your/repo and write the output to /path/to/output". You get an evidence-cited readiness
   report. The audit never writes inside your repository.
3. Say: "Start implementing the fixes on /path/to/your/repo based on /path/to/output/.json". The skill fixes
   the confirmed findings one theme at a time, on local branches. It never merges.

No clone wanted? Download the two skill zips below, check them with SHA256SUMS, and unzip them into your project's
.claude/skills/ folder: install the skills.

What changed since v0.2.0

• One layout: docs/, spec/, greenfield/, audit/, implement/, dev/. Old paths are listed in the
  "Moved" table of CHANGELOG.md.
• The skills start from a clone with no install step.
• Inside the audit bundle the engine moved to scripts/engine/audit/. Unzip the new release over the old folders.
• Each bundle now carries its licence (AGPL-3.0-or-later).
• SECURITY.md, a protected main, and secret scanning are in place.

Platforms

Linux: tested. macOS: the tools are tested in CI; the Claude Code path is not yet checked by hand. Windows: not
tested (WSL may work).

Files

File	What it is
agentic-readiness-audit-v0.3.0.zip	The audit skill
agentic-readiness-implement-v0.3.0.zip	The implement skill (install beside the audit skill)
audit.pyz	The deterministic engine as one file: python3 audit.pyz --help
SHA256SUMS	Check the downloads: sha256sum -c SHA256SUMS

Full notes: CHANGELOG.md.

v0.2.0

See CHANGELOG.md. Verify a download with: sha256sum -c SHA256SUMS