xberg-io/agent-conventions

github-workflow-conventions

Organize GitHub Actions workflows, name CI/release/deployment files, pin actions, and audit workflow references and stage boundaries. Use when creating, renaming, splitting, or reviewing workflows.

原文语言：英语

context-improvement

Run one round of the context improvement loop over eval runs: triage failures, one delta patch per hypothesis with placebo and deletion arms, the pre-registered accept rule, the ledger, holdout checkpoints.

原文语言：英语

mise-toolchains

mise procedures: locked install, adding a tool, moving to a new baseline, matrix version legs, lock and rate-limit failures. Use when installing or changing a tool version.

原文语言：英语

xberg-brand-and-docs

xberg.io brand, product catalog, docs site (Astro Starlight, @xberg-io/docs-theme), README template, badges, analytics IDs. Use when writing READMEs or docs, or editing docs-site config.

原文语言：英语

adr-authoring

Write, supersede or seed MADR decision records with adrlint: shape, Confirmation, lock, index. Use when adding or changing an ADR or an adrs/ log.

原文语言：英语

agent-coordination

Dispatch, monitor, recover and verify subagents and worktree peers in a shared repo. Use when spawning a subagent, checking a quiet one, merging an agent's branch, or writing a brief for another agent.

原文语言：英语

e2e-generation

Fixture-driven cross-language e2e suites: generated e2e/ output, the [crates.e2e] config, fixture files and ids, adding a language, regenerate and verify. Use when adding or editing fixtures or configuring e2e generation.

原文语言：英语

release-workflow

Cut a release: version task, CHANGELOG, clean-tree gate, push then tag, gh release create, check the publish trigger, install the released build. Use when cutting or publishing a version.

原文语言：英语

mise-tasks

Run and write tasks with mise: mise run, ns:verb names, mise.toml vs mise-tasks/ file tasks, usage args, depends, root-task guard, CI. Use before a raw build, test, lint or format command, or when adding a task.

原文语言：英语

scoop-distribution

Windows Scoop channel: manifest templates in scripts/publish, the publish-scoop-manifest job, the shared bucket, adding a release target. Use when editing Scoop manifests, adding a CLI or release target, or debugging a Scoop install.

原文语言：英语

basemind-tools

Which basemind plugin skill to load for code search, history, documents or agent coordination, plus the graph modes and gotchas they omit. Use when navigating code or history in a repo where basemind is available.

原文语言：英语

binding-architecture

Org layout for polyglot bindings: crate and package naming, distribution paths, what the generator decides vs the architect, binding review checklist, per-language doc formats. Use when designing, adding or reviewing a binding.

原文语言：英语

feature-flag-conventions

Shared Cargo feature-flag names and semantics across the Rust libraries (otel, mcp, mcp-http, mimalloc, jemalloc, acceleration). Use when adding, renaming or reviewing a crate's [features].

原文语言：英语

poly-lint-format

poly, the multi-language linter and formatter: engines, lint and fmt commands, poly.toml, severity, the uncomment pass, CI workflow. Use when linting, formatting or configuring poly.toml.

原文语言：英语

tdd-workflow

Test-driven workflow: red, green, refactor; restart if code came first; reproduce a bug as a failing test; integration, unit or property test by change type. Use before a feature or bug fix or when choosing a test type.

原文语言：英语

tracing-conventions

Rust tracing contract: level semantics, span and field naming, emit-vs-install, instrumentation checklist. Use when adding spans or events, wiring a subscriber, or reviewing instrumentation.

原文语言：英语

verification-harness

Zero-work passes seen in practice and how to guard a harness against them: xargs limits, empty path variables, wrong-reason negative controls, job-count assertions. Use when writing or trusting a sweep, gate or check script.

原文语言：英语

consumer-release-gates

Reproduce a generated-bindings repo's release gate locally and verify published package names against the real manifests. Use before tagging a release or when auditing a registry publish.

原文语言：英语

code-quality-standards

Code-quality limits: 120-char lines, complexity 20, nesting 4, 50-line functions, no magic numbers or global state, rule of three, delete dead code, narrow suppressions, tracked TODOs. Use when writing, reviewing or refactoring code.

原文语言：英语

config-placement

Where an instruction belongs in AI-assistant config: always-on root file, path-scoped rule, directory context file, skill or agent, and how to review one. Use when adding or reviewing a rule, skill, agent or context file.

原文语言：英语

dependency-awareness

Dependency vulnerability audits per language: cargo audit/deny, pip-audit, pnpm audit, govulncheck, bundler-audit, composer audit, dotnet, mix_audit. Use when adding or updating dependencies or triaging CVEs.

原文语言：英语

error-handling

Error handling by language: wrap with context, never swallow, Result in Rust, %w wrapping in Go, typed exceptions in Python and Java, fail fast. Use when writing or reviewing an error path or custom error type.

原文语言：英语

incremental-approach

Build in verified increments: smallest change first, verify, then extend; prototype the risky part first; batch edits to one file. Use when planning a large or multi-file change or choosing between approaches.

原文语言：英语

owasp-quick-reference

OWASP Top 10 with one-line mitigations: access control, crypto, injection, insecure design, misconfiguration, vulnerable components, auth, integrity, logging, SSRF. Use when reviewing code or a design for security.

原文语言：英语

skill-authoring

Create or improve a skill: decide whether a skill is the right home, interview the user, write a brief SKILL.md, test it on real tasks and iterate. Load when asked to create a skill, turn team knowledge into one, or move a command into one.

原文语言：英语

scoop-distribution

Maintain the Windows Scoop release channel: the per-repo manifest template and config, the publish-scoop-manifest job, and the shared xberg-io/scoop-bucket. Use this skill when editing anything under scripts/publish/*.json.tmpl or scripts/publish/scoop.json,…

原文语言：英语

agent-coordination

How to dispatch, monitor, recover, and trust subagents and worktree-isolated peers safely in a repo other agents may also be touching. Use this skill whenever you spawn a subagent, check on one that has gone quiet, merge an agent's branch, act on a subagent's…

原文语言：英语

basemind-tools

How to use basemind's MCP tools and CLI for structural and historical code questions instead of grep, file reads, and naked git. Load when navigating the codebase, finding definitions or call sites, exploring git history, or coordinating with other agents in…

原文语言：英语

release-workflow

Release/publish a Rust core crate or CLI in an xberg-io polyglot repo (crawlberg, html-to-markdown, xberg, tree-sitter-language-pack, liter-llm). Load when releasing or publishing a Rust crate or CLI — cutting a new version, tagging, running `gh release…

原文语言：英语

poly-lint-format

Reference for poly, the single-binary multi-language linter and formatter — per-language engines, the lint/fmt commands, poly.toml configuration, severity behavior, and the shared CI validation workflow. Load when linting or formatting code, configuring…

原文语言：英语

gcloud-conventions

Google Cloud conventions for xberg-io — project and region, the standard service set, resource naming, workload identity federation over service account keys, least-privilege IAM, and gcloud-CLI-only operations. Load when provisioning or configuring GCP…

原文语言：英语