ADLC5 5.3.1 public-preview candidate

ADLC5 5.3.1 public-preview candidate packages for Codex, Claude Code, Cursor and Hermes Agent.

Choose the archive for your agent. ZIP and tar.gz contain portable source, not a bundled interpreter or agent. Install Python 3.10+, Git, Bash 3.2+ and jq. Extract to an empty folder, inspect INSTALL.json for provenance, then run python3 install.py --dry-run and python3 install.py (Windows PowerShell: py -3 install.py). Complete the host-specific steps printed by the installer. Unix users should prefer tar.gz to preserve executable bits.

The same archives are also published to GitHub Packages as ghcr.io/dango-tech/adlc5-<host>-plugin:5.3.1.

What's new since 5.2.0

5.3.1 ships the 5.3.0 changes; the v5.3.0 release has no packages because its upload failed on an immutable release. The release workflow now attaches packages to a draft first (#42).

• Pulse shows real token use (#36): when a feature has no hook ledger, the board and terminal band fall back to transcript token totals from matched sessions, labeled as transcript-based and unpriced.
• Pulse workflow summary and links (#36): spec, stories, delivery, checks and review notes per feature; clickable cards; Codex session links; file links on the local all-projects board only.
• adlc5 config set (#36) with --scope machine|project|worktree.
• Token-burn guidance (#36): build each story in a fresh subagent, start each lifecycle in a fresh thread, and wait with one long blocking wait instead of short polls.
• Duplicate policy keys (#36) are named with both lines instead of a bare "duplicate YAML key".
• High-risk file anchors (#37, #39, fixes #31): anchors lock fails early with file_anchor_required when a high-risk feature has no file-backed criterion, and a human-approved anchor_addition_approval adds one after Implement starts.

Changelog · Installation and host behavior · OS prerequisites

Source: e8bc90a. This is a prerelease, not a stable production certification. Native host commands are mocked in installation contract tests; these results are not signed-in end-to-end qualification across every OS/host combination.

ADLC5 5.2.0 public-preview candidate

ADLC5 5.2.0 public-preview candidate packages for Codex, Claude Code, Cursor and Hermes Agent.

Choose the archive for your agent. ZIP and tar.gz contain portable source, not a bundled interpreter or agent. Install Python 3.10+, Git, Bash 3.2+ and jq. Extract to an empty folder, inspect INSTALL.json for provenance, then run python3 install.py --dry-run and python3 install.py (Windows PowerShell: py -3 install.py). Complete the host-specific steps printed by the installer. Unix users should prefer tar.gz to preserve executable bits.

The same archives are also published to GitHub Packages as ghcr.io/dango-tech/adlc5-<host>-plugin:5.2.0.

What's new since 5.1.0

• Agent-product governance (#30): product_type: agent adds the agent_governance check to plan-complete, requiring design/1d-agent-governance.md. Plan, QA, Specify and Discover gain agent-specific steps.
• GitHub Packages publishing (#28): each release's four host archives are pushed to GHCR and pulled back for checksum verification.
• Pulse band after entering a worktree (#29) shows the worktree's workspace instead of the main checkout's.
• Skill registry lists all skills, with a contract test against drift.
• No duplicate hooks in the ADLC5 repo (#32): the classic Claude hook template moved to templates/claude-settings.json; --with-hooks --hooks-platform claude now also works from a plugin package.

Changelog · Installation and host behavior · OS prerequisites

Source: 11528ee. This is a prerelease, not a stable production certification. Native host commands are mocked in installation contract tests; these results are not signed-in end-to-end qualification across every OS/host combination.

ADLC5 5.1.0 public-preview candidate

ADLC5 5.1.0 public-preview candidate packages for Codex, Claude Code, Cursor and Hermes Agent.

Choose the archive for your agent. ZIP and tar.gz contain portable source, not a bundled interpreter or agent. Install Python 3.10+, Git, Bash 3.2+ and jq. Extract to an empty folder, inspect INSTALL.json for provenance, then run python3 install.py --dry-run and python3 install.py (Windows PowerShell: py -3 install.py). Complete the host-specific steps printed by the installer.

Installation and host behavior · OS prerequisites

Validation run: all seven archive/installer tests passed on ubuntu-latest, macos-latest and windows-latest. Tests cover all four archives, checksum/provenance, write-free dry runs, extracted ZIPs in paths with spaces, supported host command plans and preservation of existing destinations. Native host commands are mocked in installation contract tests; these results are not signed-in end-to-end qualification across every OS/host combination. Unix users should prefer tar.gz to preserve executable bits.

Includes requirements readiness, central Pulse and Settings, configuration layers, and self-install packages merged in PR #26. Documentation synchronization is in PR #27. Source: df1ee33. This is a prerelease, not a stable production certification. The comparative campaign remains unexecuted.