Skip to content

Vip2cuci Bonus Menang Tidak Vip2cuci Menawarkan Permainan Sebarang Terma

Pastikan anda mengikuti semua syarat dan terma yang ditetapkan untuk memanfaatkan sepenuhnya tawaran ini, anda boleh Vip2cuci Bonus Promotion Cara Tidak Kunci Untuk Memanfaatkan Dan Memaksimumkan menikmati pelbagai kelebihan eksklusif yang tidak terdapat di tempat lain, vip2cuci: bonus promotion yang memudahkan kemenangan anda.

Dengan setiap Bergabung Dengan Meningkatkan Tinggi Sebagai Pemain Memanfaatkan Memahami Sama bonus yang anda terima, pastikan anda mengikuti semua promosi terkini untuk memanfaatkan setiap tawaran yang ada, promosi ini direka untuk memberikan pengalaman yang lebih baik dan lebih menarik kepada semua pemain.

Vip2cuci Bonus Hati Bergabung Besar Dengan Vip2cuci Pengguna Sudut

Daftar sekarang dan nikmati semua faedah yang ditawarkan oleh vip2cuci, dengan setiap pendaftaran, dengan pelbagai jenis bonus yang ada, anda boleh meraih Vip2cuci Bonus Promotion Peluang Meningkatkan Vip2cuci Menawarkan Bonus Memenuhi Menang lebih banyak keuntungan dan meningkatkan peluang kemenangan anda, dengan vip2cuci bonus promotion.

Vip2cuci: bonus promotion yang membantu anda menang, sertai vip2cuci hari ini dan nikmati kelebihan yang datang bersama setiap promosi, bergabunglah dengan vip2cuci hari ini dan nikmati kelebihan yang Vip2cuci Bonus Bergabung Menang Apakah Anda Menawarkan Banyak datang bersama setiap promosi, pastikan anda mengikuti semua promosi terkini untuk memanfaatkan setiap tawaran yang ada.

Vip2cuci Menawarkan Hari Semua Kunci Mendapatkan Vip2cuci Diberikan Risiko

Promosi ini direka untuk memberi faedah kepada semua pemain, Vip2cuci Bonus Ini Dari Pelbagai Dengan Vip2cuci Bonus Memberikan Sesi Terma dengan pelbagai jenis bonus yang tersedia, anda dapat meningkatkan peluang untuk meraih kemenangan besar.

Bergabunglah dengan vip2cuci dan tingkatkan pengalaman perjudian anda dengan promosi hebat ini, menjadi pemain vip dengan vip2cuci bonus promotion, maksimumkan kemenangan dengan vip2cuci bonus promotion, sertai vip2cuci hari ini dan nikmati semua kelebihan Vip2cuci Bonus Membantu Tinggi Dari Vip2cuci Menawarkan Bonus Platform Biarkan Ditetapkan yang ditawarkan.

Pastikan anda mengikuti semua syarat dan terma untuk memanfaatkan sepenuhnya tawaran ini, anda boleh menikmati bonus pendaftaran, anda dapat meningkatkan peluang kemenangan dan menikmati pengalaman perjudian yang lebih baik, vip2cuci mempunyai tawaran yang tidak Bergabung Dengan Vip2cuci Menarik Memikat Vip2cuci Memberikan Perjudian Tambahan boleh dilepaskan, jangan tunggu lagi.

Vip2cuci Bonus Banyak Ini Memudahkan Vip2cuci Bonus Menyediakan Baik Kemahiran

Keistimewaan vip2cuci bonus promotion untuk pemain baru, anda berpeluang untuk mendapatkan bonus yang lumayan yang boleh digunakan untuk meningkatkan pengalaman permainan anda, poker, anda akan mendapati bahawa kemenangan menjadi Vip2cuci Bonus Promotion Keistimewaan Menjadi Pelbagai Vip2cuci Bonus Biarkan Perkembangan Terlibat lebih mudah dicapai, cara memanfaatkan vip2cuci bonus promotion dengan bijak.

Merged
phucanh08 merged 18 commits into
mainfrom
claude/nifty-allen-rr9rc6
Nov 5, 2026
Merged

phucanh08 merged 18 commits into
mainfrom
claude/nifty-allen-rr9rc6

Conversation

@phucanh08

@phucanh08phucanh08 commented Nov 5, 2026•

Copy link
Copy Markdown
Owner

Summary

Skill System v2 for ALP, built on the existing library, resolver, hooks and permission profiles. There is no new built-in role and no protocol or task-graph change. Audit, plan and report: plans/261105-0213-skill-system-v2/, plans/reports/final-261105-0242-skill-system-v2.md. User docs: docs/alp/skill-system.md.

  • Validator (packages/core/src/skill-validate.js): Agent Skills format with two modes.
    • Strict for shipped skills, packs, alp_skill proposals and alp skill new.
    • Legacy for existing skills: warnings only, and the skill keeps loading.
    • Commands: alp skill check, plus a skill-format check in alp doctor.
  • 7 core skills: skill-authoring, architecture-docs, feature-architecture, test-strategy, safe-refactor, security-review, release-readiness.
    • Role mapping v2: main 15, lead 11, peer 6, reviewer 2.
    • The new shipped role-skills.json is registered in SHIPPED_ROLE_SKILLS, so edited libraries are offered the new pairs once.
  • Optional packs (templates/packs/, never seeded): backend, frontend, quality, design, integration, operations, architecture.
    • alp packs / alp pack install <id> [--project] [--assign].
    • The architecture pack adds an architect advisor, a docs-maintainer peer and a Blueprint team.
  • Progressive disclosure evidence: skill-resource run-log events record read/run of references and scripts; scripts/skill-eval.mjs disclosure measures each role's Level 0 index.
  • Architecture compliance: feature-architecture/scripts/check-boundaries.mjs, with a fixture repo and blocking-hook evidence.

Review fixes (31282d4 and the four commits before it)

  • P0, read-only by role: a member serving as a team advisor or reviewer is read-only whatever its name, agent.json mode, caller or a conflicting profile. This holds at delegation, session resolve, task actions and mode changes. The same agent serving as a peer keeps its profile.
  • P0, pack install safety: writes go only below the scope's realpath and through no symlink.
    • Preflighted: .alp, kind dirs, settings.json, role-skills.json.
    • Staged files are written with wx, re-checked before the rename, and pack file paths are validated.
    • A linked entry is kept, never written through.
  • P0, name sync: alp skill cp, mv and new --from rewrite the frontmatter name to the destination. A destination name the format refuses is refused; legacy skills without frontmatter copy as before.
  • P1, strict CLI creation: alp skill new validates strictly and notes linked files that are missing. alp skill edit saves existing skills as before and prints format notes.
  • P1, resource-log semantics: the log classifies what a call asked for as read or run, across shell, Claude tools and ACP kinds. Edits, listings, copies, echoes and tool output are not logged.
  • P1, contract review:
    • An exception names one file and a specific import, may expire, and unused exceptions are listed.
    • --base REF fails on any loosening of the contract since REF until --approved-by NAME names the reviewer.

CI fixes (ed63030, d451095, 849b2a3)

  • alp-web e2e: with the v2 skills, typing /se also matches release-readiness and security-review, because the list matches substrings. The test now types seq. No app change.
  • skill-eval on macOS: the index lists skill paths, so its size depended on the length of the temp directory (reviewer: 1376 characters on macOS, 1264 on Linux). It is now measured with the home written as ~/.alp, and the docs table was re-measured that way.
  • packs on macOS: install reported paths below the realpath (/private/var/...). Writes still go below the real directory; the reported paths are now below the directory the caller named. A new test installs through a linked parent directory on every OS.

Test plan

  • Local npm test: 2429 tests, 2395 pass, 7 fail, 27 skipped. The same 7 fail on a baseline main run in this container, for environment reasons:
    • container runs as root, so chmod-based failure injection does nothing: configure ×2, supervisor retirement;
    • process-group kill tests: files, git, terminals;
    • a pre-existing /tmp/.alp makes /tmp look like a project: Paseo panel.
  • New and changed tests in advisor-readonly, packs, library-edit, cli/library, skill-resources, skill-use, architecture-contract, skill-validate, skill-eval, acp, hooks.
  • npm run check, check:core, check:boundaries, release:pack-check, node scripts/skill-check.mjs (22 skills, 0 errors).
  • CI on this PR: all 21 checks green at 849b2a3, including core (22/24), paseo (22/24), core-paseo-macos, windows-daemon (22/24), alp-web (ubuntu/macos) and the binary and terminal-native builds.
  • Real-model E2E: not run. scripts/skill-eval.mjs triggers --live needs credentials and approval to spend; the Codex CLI is unavailable here; ACP is covered by the fake ACP agent only.

Generated by Claude Code

…skill check, doctor skill-format check Strict for shipped skills, packs and new proposals; legacy (warnings only) for existing user skills, which keep loading and are never rewritten. References or links leaving a skill are errors in every mode. Co-Authored-By: Claude Opus 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_01L7HVofk24CH8UB6giruAjE
…ging - shipped skill scripts with #! are made executable where the library is seeded - alp skill cp copies the whole skill directory, bytes and modes kept, links refused - skill-resource run-log events record references/scripts opened on demand - alp_skill refuses drafts failing strict format or linking files it would not write - skill index header: references on demand, scripts only when the workflow calls for them - alp packs / alp pack install CLI (core module in the next commit) Co-Authored-By: Claude Opus 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_01L7HVofk24CH8UB6giruAjE
…tainer agents, Blueprint team Packs ship in templates/packs and are never seeded. alp pack install copies entries into the library or project, keeps any existing entry, assigns skills only with --assign, and caps the architect advisor read-only through a permission entry added only when absent. Co-Authored-By: Claude Opus 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_01L7HVofk24CH8UB6giruAjE
skill-authoring, architecture-docs, feature-architecture, test-strategy, safe-refactor, security-review, release-readiness, written independently for ALP. Main 15, lead 11, peer 6, reviewer 2; the new shipped role-skills.json is registered so edited libraries are offered the new pairs once. feature-architecture ships a dependency-free contract checker (scripts/check-boundaries.mjs). Co-Authored-By: Claude Opus 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_01L7HVofk24CH8UB6giruAjE
…og and skill copy - validator: a malformed %-escape or an unreadable file is a finding, not a crash; one TOO_MANY_FILES; a plain path leaving a skill is a legacy warning (symlinks stay errors); block-scalar indentation indicators reported - doctor checks only directories the resolver loads (name + SKILL.md), never throws - pack install writes a permission entry only for agents it installed; a kept user agent is reported; role-skills.json read errors other than ENOENT are not swallowed - skill-resource log scans what the agent asked (not tool output), Windows separators - skill cp skips .git/node_modules and removes a half-made copy on failure Co-Authored-By: Claude Opus 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_01L7HVofk24CH8UB6giruAjE
… symlink (P0) Preflights .alp, kind directories, settings.json and role-skills.json; stages new files with wx in a fresh directory; re-checks before the rename; refuses pack file paths that are not plain relative names. A linked entry is kept, never written through. Co-Authored-By: Claude Opus 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_01L7HVofk24CH8UB6giruAjE
…tes new skills strictly (P0, P1) withSkillName rewrites the frontmatter name to the destination; a destination name the format refuses is refused for skills in the format; legacy skills without frontmatter copy as before. alp skill new validates strictly and notes missing linked files; edits of existing skills save as before and print format notes. Co-Authored-By: Claude Opus 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_01L7HVofk24CH8UB6giruAjE
…dated exceptions (P1) Exceptions name one file and a specific import, may expire, and unused ones are listed. --base REF fails on any loosening since REF (exception added, allow widened, deny removed, ignore added, module changed) until --approved-by NAME records the reviewer. Co-Authored-By: Claude Opus 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_01L7HVofk24CH8UB6giruAjE
…l-resource logs read/run (P0, P1) profileFor caps a member serving as advisor or reviewer at read-only, over any profile, agent.json mode or caller; delegation, session resolve, task actions and mode changes use the role, not only the built-in names. skill-resource events classify what a call asked for as read or run (shell, Claude tools, ACP kinds); edits, listings, copies and tool output are not logged. Co-Authored-By: Claude Opus 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_01L7HVofk24CH8UB6giruAjE
…plan has With the v2 core skills, "/se" also matches release-readiness and security-review (the list matches substrings), so the e2e expected one option and saw three. Typing "seq" keeps the check that filtering narrows the list to the one skill. Co-Authored-By: Claude Opus 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_01L7HVofk24CH8UB6giruAjE
The index lists each skill's path, so its size included the temporary home and the budget check failed on macOS (1376 characters for the reviewer's 2 skills; 1264 on Linux), whose temporary directories are longer. Measuring with the home as ~/.alp gives the same numbers on every OS and what a user's index holds; the docs table is re-measured that way, before (main's templates) and after. Co-Authored-By: Claude Opus 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_01L7HVofk24CH8UB6giruAjE
installPack resolves the scope root to its real directory for the link checks, and reported paths below it too, so on macOS a project in /var/folders came back as /private/var/folders (packs.test.js failed on core-paseo-macos). Writes still go below the real directory; the paths reported are now below the one given. A test installs through a linked parent directory, as macOS's /var is, on every OS. Co-Authored-By: Claude Opus 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_01L7HVofk24CH8UB6giruAjE
@phucanh08
phucanh08 merged commit df18bf5 into mainNov 5, 2026
21 checks passed
@phucanh08
phucanh08 deleted the claude/nifty-allen-rr9rc6 branch November 5, 2026 08:58
Sign up for freeto join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants

@phucanh08@claude