# Changelog

## 1.3.0

See [release notes](docs/release-1.3.0.md) for installation, profile transitions, and recovery.

- Add the optional `light` skill profile in the CLI and PowerShell installers. It selects one compact `project-notes` skill without bundled references, with focused record reads and result-focused replies subject to required disclosures and host instructions.
- Reconcile unselected current AAPB skills during explicit profile migration, preserving ownership checks, local edits, unrelated plugins, previews, and recovery. Ordinary installation remains additive.
- Require valid selected replacements before migration cleanup. Omit removal plans on selected-installation conflicts and recheck all replacements before each removal, preserving remaining old copies if installation fails or changes during apply.
- Document light installation, profile transitions, and the independent minimal bootstrap choice in English and Korean. Light does not alter model settings, permissions, or audit records.

## 1.2.2

- Guide interactive setup through folder preparation, explicit choices, repository checkboxes, search/rescan, back navigation and an editable final review. Show recommendations separately from saved settings and explain unavailable local Git exclusions.
- Improve English/Korean menu wording and readable setup results while preserving argument, JSON and `--yes` automation behavior.
- Exclude worklog guides and summaries consistently in record search, including custom worklog folders.
- Match record path filters to Windows filesystem casing and keep AST protected directories excluded through case aliases.
- Align worklog search and listing on filename, folder and metadata dates while retaining bounded scans and pruning other months.

## 1.2.1

- Resolve Windows short path aliases before checking exclusion boundaries, preventing a configured global ignore file from overlapping Git metadata, project records, instructions or another exclusion scope.
- Preserve link rejection, support new ignore-file suffixes, and accept equivalent paths in existing bootstrap recovery journals.
- Correct Windows fixture path comparisons so editor-conflict injections and worktree assertions exercise the intended filesystem paths.

## 1.2.0

- Add interactive and argument bootstrap with single/workspace selection, minimal/standard records, optional AGENTS links, and local/shared/global/no exclusions.
- Support record creation without Git, preserve existing content and tracking, and return guarded bootstrap recovery journals for changes.
- Register workspace members explicitly; resolve shared records from registered descendants and select existing member-local records without moving them.
- Add monthly worklog creation/listing, topic knowledge creation, repository inventory, and bounded record filters.
- Require a selected repository for workspace-root AST and Forge operations; retain existing AST languages and four default MCP tools.
- Update English/Korean instructions and guides while preserving user model/reasoning choices, approval requirements, and detailed evidence.

Promote the tested 1.2 candidate to the stable release and align the Python module's reported version with package metadata. See [Verification](docs/verification.md) and [Local package testing](docs/demo.md).

## 1.1.0

- Restore read-only structural source search through `ast search`, with a lazily loaded optional ast-grep engine.
- Add `aapb_ast_search` only when MCP starts with `--with-ast`; keep the four default record tools and existing skill profiles.
- Bound source selection, parsing, snippets and result pages; report incomplete coverage and reject changed-source cursors or linked paths.
- Document language selection, shell quoting, npm engine installation and one-time MCP activation in English and Korean.

## 1.0.0

- Provide portable project records, six selected skills, neutral project instructions, and optional read-only record MCP tools.
- Offer `ai-agent-playbook` as the primary command and `aapb` as its short alias; omitted project paths use the current directory.
- Preserve user edits during installation and migration, provide recovery journals, and keep Forge coordination explicit.
- Retire the previous executor, scheduler, and duplicate analysis commands with pinned 0.5.11 recovery guidance.
- Include complete English/Korean usage, migration, architecture, and release documentation in npm packages.

[Release notes](docs/release-1.0.0.md)

## 1.0.0-next.3

- Replace stack-specific root instruction templates with one neutral template, preserving domain contracts in selected references and explaining project-owned architecture decisions.
- Restore useful personal instruction detail and correct obsolete bootstrap references without changing automatic installation behavior.

- Add `ai-agent-playbook` as the primary executable and retain `aapb` as an equivalent short alias.
- Document and test the existing current-directory default when a project argument is omitted.
- Expand the command guide with complete invocations, option meanings, write behavior, and recovery examples in English and Korean.
- Replace implementation-design notes with a reader-facing 1.0 rationale and 0.5.11 guidance; explain skill selection and MCP verification separately.

## 1.0.0-next.2 (source prerelease)

- Restore README branding and localized navigation; rebuild beginner, command, recovery, integration, and maintenance guides with worked examples.
- Include README images, context, and examples in npm archives; deactivate examples that referenced removed hooks and remove orphaned retired-skill metadata.
- Rename the four MCP tools to aapb_status/search/read/validate; keep package and record-directory names.
- Replace clipped JSON previews with adjustable content budgets and source-bound continuation.
- Preserve source text, match locations, validation totals and incomplete scan evidence across pages.
- Separate the content budget from a 256 KiB complete-result ceiling and document prerelease allowlist migration.
- Allow up to eight seconds for Python interpreter probes and include candidate diagnostics when validation cannot find a usable engine.

## 1.0.0-next.1 (source prerelease)

- Replace broad process routing with selected project-record, artifact, design, UI, prose and optional legacy-contract skills.
- Use one default installation root with migration journals and guarded rollback.
- Keep default backups beside custom installation roots and reject cross-filesystem backup transactions during preview.
- Reject empty explicit skill selections and unreadable current-state entrypoints before migration.
- Honor the writing report root and reject linked ancestor paths in advisory checks.
- Cap the complete MCP tool result and include Korean documentation in npm archives.
- Reduce MCP to four project-bound record tools and preserve legacy record reading.
- Retire autonomous execution, scheduling, automatic delivery and duplicate analysis. Keep explicit forge coordination.
- Registry publication and stable promotion are separate from this source version.

This file summarizes shipped, user-visible changes. Current behavior belongs in the feature and operator documents under `docs/`; detailed release artifacts remain on [GitHub Releases](https://github.com/jukrap/ai-agent-playbook/releases).

## 0.5.11 (2026-08-15)

- Added explicit preserve, managed-link, and guarded-replace modes for repositories that already contain `AGENTS.md`.
- Preserved existing `.gitignore` encoding and line-ending choices while adding only the missing local playbook pattern.
- Added read-only generic UI candidate scanning and before/after writing-fidelity evidence.
- Tightened Korean prose review around repetition, density, context, and protected facts.

[Release details](https://github.com/jukrap/ai-agent-playbook/releases/tag/v0.5.11)

## 0.5.10 (2026-07-19)

- Aligned hosted GitHub/Gitea workflow package pins and Forge request identity with the shipped package version.
- Made Python interpreter discovery continue after an isolated process-creation failure.

[Release details](https://github.com/jukrap/ai-agent-playbook/releases/tag/v0.5.10)

## 0.5.9 (2026-07-12)

- Separated retry-budget usage from monotonic attempt identity so a resumed run cannot reuse append-only ledger event IDs.

[Release details](https://github.com/jukrap/ai-agent-playbook/releases/tag/v0.5.9)

## 0.5.8 (2026-07-12)

- Filtered provider-confirmed no-op reconciliation work before apply.
- Reused compatible Project fields, items, and the system-created default view without repeated writes.
- Added strict migration of legacy managed issue-body regions while preserving user-authored text.

[Release details](https://github.com/jukrap/ai-agent-playbook/releases/tag/v0.5.8)

## 0.5.7 (2026-07-12)

- Made interrupted issue supersede flows recoverable with CAS-safe ordering and approved-marker discovery.
- Removed obsolete Project items without deleting the underlying issues or user content.

[Release details](https://github.com/jukrap/ai-agent-playbook/releases/tag/v0.5.7)

## 0.5.6 (2026-07-12)

- Added actionable GitHub Projects permission recovery before the first remote mutation.
- Replaced tool-branded Project field defaults with neutral delivery fields while retaining legacy aliases.
- Corrected user-owned Project View requests to use the owner login expected by GitHub's public API.

[Release details](https://github.com/jukrap/ai-agent-playbook/releases/tag/v0.5.6)

## 0.5.5 (2026-07-12)

- Separated fine-grained local execution tasks from human-facing roadmap and delivery-group issues.
- Added reviewable issue bodies, noun-phrase title checks, Project fields and views, and guarded legacy-issue reconciliation.
- Kept merge approval separate from implementation, verification, and draft-PR delivery.

[Release details](https://github.com/jukrap/ai-agent-playbook/releases/tag/v0.5.5)

## 0.5.4 (2026-07-11)

- Added resumable execution ledgers, single-task automation ticks, controller verification, and local supervision.
- Added capability-gated GitHub/Gitea coordination, preview-first scheduling, isolated unattended workspaces, and draft-PR delivery.
- Kept local-only execution available when remote coordination is unavailable or denied.

[Release details](https://github.com/jukrap/ai-agent-playbook/releases/tag/v0.5.4)

## Earlier releases

See [v0.5.3](https://github.com/jukrap/ai-agent-playbook/releases/tag/v0.5.3) and the [complete release list](https://github.com/jukrap/ai-agent-playbook/releases) for earlier history.
