starred/nix-config-ryan4yin
1
0
Fork 0
mirror of https://github.com/ryan4yin/nix-config.git synced 2026-08-27 05:44:00 +02:00
Code Issues Packages Projects Releases 10 Wiki Activity

docs(agents): streamline global guidance

Browse Source 
Signed-off-by: Ryan Yin <xiaoyin_c@qq.com>
This commit is contained in:
Ryan Yin
2026-08-26 20:14:18 +08:00
parent 1e5626b8b7
commit 73bf26f5e2
2 changed files with 124 additions and 107 deletions
Download Patch File Download Diff File Expand all files Collapse all files
agents/AGENTS.md
+82 -84
View File
@@ -1,110 +1,108 @@
# Personal Global Agent Rules # Personal global agent rules
This file defines my global defaults and safety boundaries for coding agents. These rules define my default safety boundaries and working preferences for coding agents.
## 1) Instruction Priority ## Scope and precedence
Apply instructions in this order: Apply instructions in this order:
1. Runtime system/developer instructions 1. Runtime system and developer instructions
2. User task request 2. The current user request
3. Project-local policy (`AGENTS.md`, `CLAUDE.md`, repo docs) 3. Safety and secret-handling rules in this file
4. These global rules 4. Project-local policy (`AGENTS.md`, `CLAUDE.md`, and repository documentation)
5. Other defaults in this file
If rules conflict, follow the higher-priority source and state the conflict briefly. Project-local policy may override global defaults but must not weaken safety or secret handling.
Follow the higher-priority source when rules conflict and state the conflict briefly.
## 2) Hard Safety Boundaries (MUST NOT) ## Safety and authorization
- MUST NOT read/write outside the approved workspace. ### Workspace access
- MUST NOT perform broad operations on the entire home directory.
- MUST NOT mutate remote Git state unless explicitly requested.
- Examples: `git push`, creating/updating remote PRs/Issues via `gh`.
- MUST NOT auto-run remote-mutating commands unless explicitly requested.
- Examples: `kubectl apply/delete`, `helm upgrade`, `terraform apply`, remote `ssh` mutation.
- MUST NOT perform destructive or irreversible operations, or use force options (e.g. `rm -rf`,
`terraform destroy`).
- MAY perform explicitly requested, verified, and recoverable cleanup (e.g. `git branch -d` for a
fully merged branch).
- MUST NOT expose or commit secrets (tokens, keys, kubeconfig credentials, passwords).
## 3) Security and Secrets Handling - Access only runtime-approved roots and paths explicitly placed in scope.
- Do not perform broad operations on the entire home directory.
- Never write secret literals into tracked files. ### Remote changes
- Use environment variables, secret managers, or placeholders.
- Redact sensitive output in logs and summaries.
- For infra/IaC changes, prefer plan/eval/check before apply/deploy.
### Secret Access - Do not mutate remote state unless the user explicitly requests it.
- Remote mutations include `git push`, creating or updating remote PRs and Issues via `gh`,
`kubectl apply/delete`, `helm upgrade`, `terraform apply`, and remote `ssh` changes.
- For infrastructure and IaC changes, prefer plan, eval, or check commands before applying or
deploying.
- When explicitly requested, an authentication client or command MAY consume a user-designated ### Destructive and force operations
secret source solely to authenticate to the specified service (e.g. an API, `redis-cli`, `psql`,
or `pgcli`).
- Secrets MUST remain opaque to the agent and must not be exposed in arguments, output, or logs;
copied, cached, or persisted; or sent anywhere except the intended authentication target.
- All other secret-value access is forbidden. Metadata and identifiers MAY be queried only with
operations verified not to reveal secret values, such as `kubectl describe secret`.
## 4) Scope Discipline - Avoid irreversible operations.
- Use destructive or force options only when the user explicitly requests or approves them, the
exact target and scope have been verified, and a recovery path or safety guard is available.
- Prefer recoverable alternatives and safeguards such as `git branch -d` and
`git push --force-with-lease`.
- Keep changes strictly within requested scope. ### Secrets and authentication
- Do not refactor unrelated areas unless user asks.
- Preserve backward compatibility unless a breaking change is explicitly requested.
## 5) Change Hygiene - Never expose or commit tokens, keys, passwords, kubeconfig credentials, or other secrets.
- Never write secret literals into tracked files. Use environment variables, secret managers, or
placeholders.
- Redact sensitive values from command output, logs, and summaries.
- When explicitly requested, an authentication client may consume a user-designated secret source
only to authenticate to the specified service.
- Keep secrets opaque. Do not expose them in arguments or output, copy them, cache them, persist
them, or send them anywhere except the intended authentication target.
- Do not access secret values for any other purpose. Query metadata or identifiers only with
commands verified not to reveal values, such as `kubectl describe secret`.
- Keep diffs minimal and reviewable. ## Repository state
- Group logically related edits together.
- Do not revert user/unrelated changes unless explicitly asked.
- Do not claim verification you did not run.
## 6) Tooling Defaults At the start of repository work:
- Use Nushell for personal tooling, Bash for simple one-off commands, and Python for scripts or - Fetch `origin` when available and use its latest default branch as the baseline.
complex control flow. - If local history differs, ask which state to use before editing.
- Prefer structural search tools first for code find/replace (`ast-grep`/`jq`/`yq`), then text tools
(`rg`, `fd`).
- Prefer project task runners (`just`, `make`, `npm scripts`, etc.) over ad-hoc commands when
equivalent.
- On NixOS, do not assume FHS paths or conventional system/package installers: they may not work and
can pollute the managed environment. For missing tools or dependencies, use `nix run`, the
project's flake/dev shell, or its existing `uv`/`pnpm` workflow; otherwise, ask the user.
- Use `gh` CLI for authorized GitHub operations, especially code/PR/issue search and inspection.
- Keep waits bounded and observable: avoid long uninterrupted sleeps and unbounded loops.
- For waits or potentially blocking operations, use short polling intervals with progress output and
explicit timeouts or bounds; do not treat a timeout as proof that the underlying process is still
running. Prefer Python over Bash for custom polling logic.
- For subprocesses that may fork or daemonize, verify that inherited stdout/stderr pipes cannot keep
the parent blocked; avoid unconditional `capture_output` when output is not needed.
## 7) Environment Defaults ## Change discipline
- Primary OS: NixOS & macOS. - Keep work within the requested scope. Do not refactor unrelated areas unless asked.
- Shell: default to Nushell, Bash also exists. - Preserve backward compatibility unless the user explicitly requests a breaking change.
- Common project locations: - Keep diffs minimal, reviewable, and grouped by logical purpose.
- Open-source project sources: `~/codes/src/` - Do not revert user changes or unrelated changes unless explicitly asked.
- Personal Git repositories: `~/codes/` - Verify changes in proportion to their risk. Never claim a check passed unless it was run.
- Work-related projects: `~/work/`
- These paths are location hints only; access them only when they are explicitly in scope or
approved.
## 8) Script Engineering Principles ### Commit messages
Treat scripts as interruptible jobs that must be diagnosable and safe to rerun: - Follow the repository convention. Otherwise, use Conventional Commits.
- Derive the message from the staged diff and use an imperative subject no longer than 72
characters.
- Keep one logical change per commit. Do not amend commits or skip hooks unless explicitly
requested.
- Verbose logging of progress, decisions, and errors. ## Tools and environment
- Stage workflows with selective execution via cli flags.
- Idempotent reruns; persist progress and support resume.
- Cache external data with invalidation.
- Separate HTTP transport from business success; retry with backoff.
- Verify key outputs independently.
## 9) Communication Defaults - Primary platforms are NixOS and macOS.
- Use Nushell for personal tooling and Bash for simple one-off commands. Prefer Python for scripts,
complex control flow, polling, and process supervision.
- Use `rg` and `fd` for discovery, `ast-grep` for syntax-aware code work, and `jq` or `yq` for
structured data.
- Prefer project task runners such as `just`, `make`, and package scripts over equivalent ad hoc
commands.
- On NixOS, do not assume FHS paths or conventional system package installers. Use `nix run`, the
project flake or dev shell, or the project's existing `uv` or `pnpm` workflow. Ask before using a
different installation method.
- `npx` is allowed for temporary or skill-provided CLIs when it does not modify project dependencies
or lock files.
- Use `gh` for authorized GitHub operations, especially code, PR, and Issue search or inspection.
- Respond in the user's language; prefer English or Chinese. ## Commands and scripts
- Code, commands, identifiers, and code comments: Prefer English.
- Prefer native wait or subscription tools. Otherwise, poll with progress and an explicit deadline,
using intervals of a few seconds for short-lived local validation. A timeout does not prove the
process is still running.
- For processes started by the agent, track and wait on the child PID or process handle directly. Do
not infer liveness by matching `ps` or `pgrep` output.
- For long-running, batch, networked, or expensive jobs, log progress; support selective stages,
idempotent reruns, and resume; define cache invalidation; distinguish HTTP success from business
success; retry transient failures with backoff; and verify important outputs independently.
## Communication
- Respond in the user's language. If the language is unclear, default to English or Chinese.
- Prefer English for code, commands, identifiers, and code comments.
- Be concise, concrete, and action-oriented. - Be concise, concrete, and action-oriented.
## 10) Project Overlay
Project-local policy may add stricter constraints (build/test/deploy/style/ownership/environment).
It must not weaken this baseline.
agents/install-skills.md
+42 -23
View File
@@ -1,7 +1,7 @@
# Agent Skills Commands # Agent Skills Commands
Reference commands for listing, installing, and updating skills via `npx skills`. Run only the Reference commands for listing, installing, and updating skills via `npx skills`. Keep the global
commands you need. set small and install task-specific skills in the relevant project.
## Inspect and update installed skills ## Inspect and update installed skills
@@ -17,47 +17,66 @@ npx skills check
# update all installed skills # update all installed skills
npx skills update npx skills update
# remove from global scope
npx skills remove --global web-design-guidelines
``` ```
## Discover skills from repositories ## Discover skills from repositories
```bash ```bash
# list skills in a repository # list skills in a repository
npx skills add vercel-labs/agent-skills --list npx skills add anthropics/skills --list
``` ```
## Install commonly used skill packs ## Global baseline
```bash ```bash
# superpowers # structured planning, testing, debugging, review, and verification workflows
npx skills add -g obra/superpowers --skill '*' npx skills add -g obra/superpowers --skill '*'
# github skills # discover task-specific skills when needed
npx skills add -g github/awesome-copilot --skill 'git-commit' --skill 'gh-cli' npx skills add -g vercel-labs/skills --skill 'find-skills'
# find skills
npx skills add -g vercel-labs/skills
# anthropic skills
npx skills add -g anthropics/skills --skill 'skill-creator' --skill 'pdf'
``` ```
## Optional packs `find-docs` is also installed globally from a local source, so it has no upstream install command.
## Optional global skills
```bash ```bash
npx skills add -g pbakaus/impeccable --skill '*' # rewrite prose to sound natural while preserving claims and technical meaning
npx skills add -g blader/humanizer --skill 'humanizer'
npx skills add -g coreyhaines31/marketingskills --skill '*' # review code for unnecessary abstractions and over-engineering
npx skills add -g DietrichGebert/ponytail --skill 'ponytail-review'
npx skills add -g phuryn/pm-skills --skill '*' # enable a terse response mode when context or token usage matters
npx skills add -g JuliusBrussee/caveman --skill 'caveman'
```
## Optional project skills
Run these commands from the project root. They intentionally omit `-g`, so the skills apply only to
the current project. Check `git status` after installation and commit the generated files only when
the whole team should use them.
```bash
# design distinctive UI and demo pages
npx skills add anthropics/skills --skill 'frontend-design'
# test local web applications with Playwright
npx skills add anthropics/skills --skill 'webapp-testing'
# read, create, edit, or validate PDF files
npx skills add anthropics/skills --skill 'pdf'
# run CodeQL and Semgrep on repositories written in supported languages; not for pure Nix projects
npx skills add trailofbits/skills --skill 'codeql' --skill 'semgrep'
``` ```
References: References:
- https://github.com/obra/superpowers
- https://github.com/vercel-labs/skills - https://github.com/vercel-labs/skills
- https://github.com/pbakaus/impeccable - https://github.com/blader/humanizer
- https://github.com/coreyhaines31/marketingskills - https://github.com/DietrichGebert/ponytail
- https://github.com/phuryn/pm-skills - https://github.com/JuliusBrussee/caveman
- https://github.com/anthropics/skills
- https://github.com/trailofbits/skills