Ghfs

Ghfs

Overview

Use this skill to operate ghfs as a local filesystem mirror for GitHub issues and pull requests.

• ghfs sync mirrors remote content into .ghfs/.
• ghfs execute merges operations from .ghfs/execute.yml, .ghfs/execute.md, and per-item markdown frontmatter differences.
• Default execute mode is report mode. Use --run to mutate GitHub.

Key .ghfs files:

• execute.md: queued commands (human-friendly)
• execute.yml: queued operations (YAML array)
• schema/execute.schema.json: schema for execute.yml
• .sync.json: sync and execution run history (skip reading it)
• issues.md, pulls.md, repo.json: aggregated mirror views
• issues/**/*.md, pulls/**/*.md: per-item mirrors

Main Workflow

1. Sync first when local data may be stale: run ghfs sync.
2. Queue requested changes in one or more sources:
   • .ghfs/execute.md for quick command-style edits (recommended).
   • .ghfs/execute.yml for explicit structured operations.
   • .ghfs/issues/**/*.md or .ghfs/pulls/**/*.md frontmatter for direct per-item edits.
3. Validate and preview with ghfs execute.
4. Apply only on explicit user intent: ghfs execute --run.
5. Report results and remaining queued operations.

Execution behavior to remember:

• Merge order is execute.yml -> execute.md -> per-item generated operations.
• Operations run in file order.
• On --run, successful operations are removed from queued YAML/MD sources; failed and not-yet-run entries remain.
• Per-item generated operations are recomputed from current frontmatter on each run.
• After run, ghfs runs targeted sync for affected numbers automatically.

Update .ghfs/execute.md Quickly

Use one command per line:

close #123 #124
set-title #125 "Improve sync summary output"
label #125 enhancement, cli
assign #126 octocat
comment #126 "Need more context"
close-comment #127 "Closing as completed"

Rules:

• Action names are case-insensitive and aliases are accepted.
• # and // line comments plus <!-- ... --> HTML comment blocks are supported and preserved.
• Multi-target commands are supported for: close, reopen, clear-milestone, unlock, mark-ready-for-review, convert-to-draft.

Common aliases:

• open -> reopen
• closes -> close
• close-comment / comment-close / close-and-comment / comment-and-close -> close-with-comment
• label / labels / tag / tags / add-tag -> add-labels
• assign / assignee / assignees -> add-assignees
• title / retitle -> set-title
• comment -> add-comment
• ready / undraft -> mark-ready-for-review
• draft -> convert-to-draft

Update .ghfs/execute.yml Precisely

Keep root as a YAML array and include number + action for each entry.

# yaml-language-server: $schema=./schema/execute.schema.json
- number: 125
  action: set-title
  title: Improve sync summary output

- number: 125
  action: add-labels
  labels: [enhancement, cli]

- number: 126
  action: request-reviewers
  reviewers: [octocat]
  ifUnchangedSince: '2026-03-05T04:10:00Z'

- number: 130
  action: add-reaction
  reaction: heart

- number: 130
  action: add-reaction
  reaction: '+1'
  target: {kind: comment, commentId: 9876543}

Map user intent to action fields:

Rules:

• number must be a positive integer.
• ifUnchangedSince must be ISO datetime when present.
• request-reviewers, remove-reviewers, mark-ready-for-review, and convert-to-draft are PR-only.
• Keep operation order aligned with user intent because execution is sequential.
• Append operations unless user explicitly asks to replace or clear the queue.

Practical number resolution:

• Parse from filenames such as .ghfs/issues/00123-foo.md -> number: 123.
• Use .ghfs/issues.md / .ghfs/pulls.md when matching by title.

Update Per-Item Markdown Frontmatter

Edit frontmatter in .ghfs/issues/**/*.md or .ghfs/pulls/**/*.md:

• title
• state (open / closed)
• labels
• assignees
• milestone

ghfs execute compares edited frontmatter against tracked state and generates operations automatically (with ifUnchangedSince safeguards).

Run Sync and Execute via CLI

Preferred commands:

ghfs sync
ghfs sync --full
ghfs sync --since 2026-03-01T00:00:00Z
ghfs execute
ghfs execute --run

Useful flags:

• --repo owner/name when repo cannot be auto-resolved.
• --non-interactive for scripted runs.
• --continue-on-error to keep applying later ops after a failure.