Agent Feed 1.1.6: Trust Model, GitHub CLI Token, and Multi-Skill Removal

What's Changed

Added

• Published a trust model document (docs/trust-model.md) explaining external hash storage, skill trust levels, custom-skill boundaries, and GitHub token lookup order.
• Added automatic gh auth token fallback for agent-feed skill-hub so users with GitHub CLI no longer need to manually set GITHUB_TOKEN.
• Added multi-skill removal: agent-feed skills remove <name> [name ...] now accepts multiple names, and opens an interactive checkbox selector when no names are passed.
• Added direct regression coverage for asset trust drift detection, GitHub CLI token fallback priority, upgrade idempotency, atomic multi-skill removal, and legacy path compatibility.

Changed

• Expanded the usage guide with env print, env uninstall, and GitHub CLI token setup guidance.
• Documented agent-feed skills remove <name> [name ...] --path <path> while keeping the older single-name final-path argument form compatible.
• Normalized Windows path detection in install source so uv and npm installs are correctly identified on Windows.

Verification

• 109 tests passing
• uv run agent-feed check -a

Agent Feed 1.1.5: Harden Skill Hub Token Retry Flow

What's Changed

• Hardened skill-hub GitHub token retry behavior across search, preview, and download stages.
• Search rate limits now retry only the current search.
• Preview rate limits now retry only the highlighted skill preview.
• Download rate limits now retry only the already selected skills, without re-running search or selection.
• Added granular regression tests for token cancelation, retry failure, token reuse, and no accidental install paths.
• Fixed Homebrew formula update helper typing issues so the full project code gate passes cleanly.

Verification

• sh .agents/scripts/verify-agent-dev.sh docs
• sh .agents/scripts/verify-agent-dev.sh code
• uv run agent-feed check -a

Agent Feed 1.1.4 - GitHub Token Retry for Skill Hub Installs

Agent Feed 1.1.4

This patch release improves the Skill Hub install flow when GitHub blocks anonymous API access.

Fixed

• Reused the existing GitHub token retry flow when selected skill downloads hit GitHub 403 or rate-limit responses after search succeeds.
• Skill Hub now shows the same GITHUB_TOKEN and settings.github_token guidance during install-time GitHub failures as it already does during search-time failures.
• GitHub download HTTP errors are normalized into user-friendly retry guidance instead of raw HTTP exception text.

Why this matters

Before this release, Skill Hub could search successfully but fail later while downloading selected skills, leaving users at an install-blocked error. Now the CLI explains the token requirement, lets users enter and optionally save a token, reruns the search, and continues the install flow.

Agent Feed 1.1.3 - Cleaner Init Backups and Noise-Free Reference Checks

Agent Feed 1.1.3

This patch release improves the AI protocol maintenance experience by making init feedback clearer and reference checks less noisy.

Highlights

• agent-feed init now shows the exact .feed-backup/<timestamp> directory when legacy AI instruction assets are backed up.
• Reference checks now focus on active protocol assets only.
• .feed-backup/ archives are no longer treated as live protocol files during checks.
• Repository history docs and stale root-level references no longer create false-positive .agents/... path errors.
• The packaged standard template now includes the same improved reference-check behavior.

Why this matters

Before this release, backup archives and historical docs could trigger irrelevant “missing referenced path” errors, which made check and docs verification noisy. This release tightens that boundary so validation reflects the current working protocol, not migration artifacts.

Included in this release

• Version bumped to 1.1.3 across Python, npm, and project metadata.
• Added regression coverage for backup feedback and reference-check noise filtering.

Verification

This release was verified with:

sh .agents/scripts/verify-agent-dev.sh docs
uv run agent-feed check -a
uv run pytest
uv run ruff check src tests scripts
uv run mypy src tests
uv build
uvx twine check dist/*
npm run pack:dry-run

v1.1.2 beatify readme and finalize

v1.1.2 beatify readme and finalize

v1.1.1 test-auto-pulish

test-auto-pulish

Agent Feed 1.1.0 - Practical Usage Guide and AI Workflow Onboarding

Agent Feed 1.1.0

This release improves the first-time user experience and makes the Agent Feed workflow easier to adopt in real projects.

Highlights

• Added a practical Agent Feed Usage Guide at docs/usage-guide.md.
• Moved the usage guide to a more visible place in both English and Chinese README files.
• Added prompt examples for starting a new AI session with Agent Feed.
• Added guidance for bootstrapping .agents/project/ and .agents/domain/ from repository code, docs, tests, and existing project context.
• Added guidance for migrating legacy AI instructions from .feed-backup/.
• Added examples for maintaining project/domain README indexes, topic summaries, and read triggers.
• Updated package metadata to version 1.1.0.

Upgrade Notes

This release does not change the core CLI command surface. It mainly improves onboarding, documentation, and the recommended workflow for maintaining project-specific AI rules.

After upgrading, read:

• docs/usage-guide.md
• README.md
• README_ZH.md

Then run:

agent-feed upgrade
agent-feed check -a

Agent Feed v1.0.1 - npm package publishing fix

What's Changed

• Publish the npm wrapper under @yysjjd/agent-feed because npm blocks the unscoped agent-feed name as too similar to an existing package.
• Update English and Chinese README install instructions to use npm install -g @yysjjd/agent-feed.
• Keep the installed CLI command unchanged as agent-feed.
• Point the GitHub release workflow to the scoped npm package page.
• Remove source-checkout development instructions from the user-facing Quick Start.

Install

uv tool install agent-feed
# or
pipx install agent-feed
# or
npm install -g @yysjjd/agent-feed

Agent Feed v1.0.0 - Initial Public Release

Agent Feed v1.0.0 is the first public release.

Agent Feed installs a repository-owned AI development protocol built around AGENTS.md and .agents/. It helps teams turn scattered AI coding instructions into a repeatable workflow for Codex, Claude Code, Cursor, verification, review, and handoff.

What It Solves

AI-assisted development often drifts because project rules live in chat history, tool-specific files, or informal habits. Agent Feed gives the repository one canonical workflow layer:

• clear task boundaries before work expands
• decision gates before architecture or verification choices become accidental defaults
• focused context loading instead of prompt flooding
• verification and review gates before claiming work is done
• session handoff rules for long-running conversations and context compression
• thin adapters for Claude Code and Cursor while keeping AGENTS.md as the source of truth

Included In This Release

• agent-feed init for installing AGENTS.md, .agents/, and selected client adapters
• agent-feed check for validating structure, references, scripts, skills, and adapters
• agent-feed sync for regenerating client adapters from canonical protocol assets
• agent-feed upgrade for non-destructive template refreshes
• agent-feed status for checking installed protocol state
• skill index generation and drift detection
• AI asset trust checks backed by external user-level config
• standard protocol template with rules, skills, project/domain layers, session-state, and helper scripts
• migration support for repositories that already have AI instruction files
• CI coverage for tests, linting, type checking, package build, and CLI smoke checks

Install

uv tool install agent-feed
# or
pipx install agent-feed

Initialize a repository:

agent-feed init
agent-feed check
agent-feed status

Status

This is a pre-1.0 release. The core workflow is usable, but command behavior, generated template structure, and verification gates may still evolve while the protocol stabilizes.

See the README and docs for the full workflow model:

• README: https://github.com/fqmyysjjd/agent-feed
• Protocol flow: https://github.com/fqmyysjjd/agent-feed/blob/main/docs/ai-development-protocol-flow.md
• Template model: https://github.com/fqmyysjjd/agent-feed/blob/main/docs/template-model.md