agentleFS
Sign inStart free
skywatch-bskyskywatch-bsky/skywatch-agent-skills/AGENTS.md

skywatch-agent-skills

AGENTS.md6 starsChanged 4 months agoMITView on GitHub

Single source of truth for Skywatch's agent assets — investigation methodology, Osprey SML rule workflows, ClickHouse/Ozone tooling, and report conventions. This is not a traditional codebase; it is a content repository that ships the same domain material to two distribution targets. The Claude Code plugins and Polytoken pack carry the same domain content with different framing: When editing content, update both targets unless the change is target-specific. Every meaningful change gets a CHANGELOG.md entry under the relevant plugin version. Bump…

  • Reads credentials
  • Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.

    Of the AGENTS.md files we read, the 31,439 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.

  • In this repositoryOne other file it gives its agents.

The file

We split this AGENTS.md into the rules it gives an agent

2 rules. About 0 are common, 1 is distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    When editing content, update both targets unless the change is target-specific.

    line 63
  2. Very specific:

    Every meaningful change gets a CHANGELOG.md entry under the relevant plugin version.

    line 67

≈ “Common” is an estimate: at least 1 in 5 of the 40,246 AGENTS.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
AGENTS.md

Skywatch Agent Skills

Single source of truth for Skywatch's agent assets — investigation methodology, Osprey SML rule workflows, ClickHouse/Ozone tooling, and report conventions. This is not a traditional codebase; it is a content repository that ships the same domain material to two distribution targets.

Tech Stack

  • No build system — Markdown files and JSON manifests only
  • Claude Code plugins — .claude-plugin/plugin.json manifests, skill/agent Markdown
  • Polytoken pack — facet/subagent/skill Markdown with polytoken: frontmatter, hooks.json
  • MCP server — external Python (FastMCP), installed via uvx from git@github.com:skywatch-bsky/skywatch-mcp.git; no server source lives here
  • Hook scripts — Node.js (validate-label-comment.js) and Bash (load-clickhouse-env.sh)

Commands

# Validate an Osprey SML rule project (from the project root, not here)
uv run osprey-cli push-rules --dry-run

# Sync Polytoken pack into a working project
rsync -a polytoken/skills/    /target/.polytoken/skills/
rsync -a polytoken/facets/    /target/.polytoken/facets/
rsync -a polytoken/hooks/     /target/.polytoken/hooks/
rsync -a polytoken/subagents/ /target/.polytoken/subagents/
cp polytoken/hooks.json /target/.polytoken/hooks.json

# Inspect shipped Polytoken definitions (for authoring reference)
polytoken vfs ls polytoken://facets
polytoken vfs cat polytoken://facets/execute.md
polytoken vfs cat polytoken://subagents/researcher.md

Project Structure

claude-skills/                    # Claude Code plugin marketplace target
├── .claude-plugin/               # marketplace manifest (name: skywatch-skills)
└── plugins/
    ├── osprey-rules/             # SML rule authoring orchestrator (v0.3.1)
    ├── osprey-rule-investigator/ # read-only SML project analysis (v0.1.1)
    └── skywatch-investigations/  # investigation toolkit (v0.25.0)
polytoken/                        # Polytoken pack target
├── facets/skywatch.md            # front-desk routing facet
├── hooks/                        # validate-label-comment.js, load-clickhouse-env.sh
├── hooks.json                    # project hook registry
├── skills/skywatch-*/            # one dir per skill (skywatch- prefixed)
└── subagents/                    # investigator, data-analyst, osprey-rule-writer
CHANGELOG.md                      # versioned changes across all plugins

Conventions

Two targets, one source — update both

The Claude Code plugins and Polytoken pack carry the same domain content with different framing:

AspectClaude Code (claude-skills/)Polytoken (polytoken/)
Skill namebare (conducting-investigations)prefixed (skywatch-conducting-investigations)
Tool nameshort (ozone_label)fully qualified (mcp__skywatch-mcp__ozone_label)
Agent frontmatterallowed-toolspolytoken: block with tools/undeferred_tools
InstallClaude Code marketplacersync into project .polytoken/

When editing content, update both targets unless the change is target-specific.

Changelog discipline

Every meaningful change gets a CHANGELOG.md entry under the relevant plugin version. Bump the version in claude-skills/.claude-plugin/*.json when shipping.

Boundaries

  • Safe to edit: claude-skills/, polytoken/, CHANGELOG.md
  • Never create: root-level CLAUDE.md (gitignored; only allowed inside plugin directories)
  • Never commit: credentials, real .envrc, secrets/*.env
  • Do not duplicate: the MCP server source lives in a separate repo — this repo only references it

Gotchas

  • osprey_execution_results is a wide event table with dynamic PascalCase columns. Only __action_id, __timestamp, __error_count, __atproto_label, __entity_label_mutations, __verdicts are universal — discover live columns with DESCRIBE TABLE, never assume flat columns like did or handle.
  • The investigator agent never writes SQL — it delegates all ClickHouse queries to the data-analyst subagent.
  • Label writes are hook-enforced: validate-label-comment.js blocks ozone_label calls unless the comment meets labeling-standards (summary line + Evidence section with AT-URI citations).
  • Ozone auth goes through the PDS (via atproto-proxy header), not directly to the Ozone service URL.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

Check your setup, free