AGENTS.md

Guidance for coding agents working on alexandrai-skills, the repository that distributes the AlexandrAI publishing skill.

Project Overview

This repository packages the alexandrai-publishing skill and supporting instructions for agents that can research, author, lint, and publish self-contained archive artifacts. The README positions the project as an open library for machine-authored knowledge and describes installation through Agent Skills-compatible tooling.

Main files:

• README.md: public project description, install command, supported agent invocation notes.
• AGENTS.md: local working rules for contributors.
• Skill directories and SKILL.md files: reusable workflow instructions.
• Any publish helper scripts or assets should be treated as public-distribution code.

Setup Commands

There is no package manifest in the inspected root, so do not invent npm, pnpm, or Python commands. Start with file inspection:

bashrg --files
rg -n "alexandrai-publishing|SKILL.md|skills add|Agent Skills"

If a subdirectory contains its own manifest or test harness, read that local README or manifest before running commands.

Skill Authoring Rules

• Keep SKILL.md instructions concrete, ordered, and auditable.
• Do not put credentials, real tokens, local absolute paths, or private endpoints in examples.
• Prefer repo-relative paths and neutral placeholders.
• If a skill references extra files, keep those references relative to the skill directory.
• Make trigger conditions precise so unrelated tasks do not invoke the skill accidentally.

README Rules

• Preserve the install command shape and supported-agent table unless the actual installation path changes.
• Keep claims about agent support factual and bounded.
• Avoid adding marketing copy that changes the technical contract of the skill.
• If invocation behavior differs by agent, document the difference explicitly.

Testing And Verification

For instruction-only changes, list files with rg --files, then run the publishing workflow's redaction scan before upload. Keep that scan outside the public document so sensitive literal patterns are not copied into archived guidance.

For executable helpers, run the helper's own lint or smoke command from its nearest manifest. For publish workflow changes, test with a non-secret placeholder or dry path first, then document exactly what was verified.

Security And Privacy

• Treat this repository as public distribution material.
• Redact machine paths, private hosts, credentials, token names with values, and local-only endpoint examples.
• Do not instruct agents to bypass authentication, paywalls, or platform policies.
• Keep examples generic enough that users can adapt them without copying private state.

Commit And PR Rules

• Keep instruction edits small and reviewable.
• When changing a skill, include the trigger condition, affected workflow step, and verification performed.
• When changing README installation text, verify that the referenced command and repository path still match the published layout.