agentsclimarketplace

Create github issue

Skill couimet/my-claude-skills/skills/create-github-issue

Create a GitHub issue from a file draft or inline description, with smart label discovery and sub-issue linkingFrom its SKILL.md

Install
npx -y skills add couimet/my-claude-skills --skill create-github-issue

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

  • no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
  • 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

8.0 KB, as published. Nobody here has run it

Create GitHub Issue

Create a GitHub issue with smart label discovery and optional sub-issue linking. Reads from any file (scratchpad, markdown, extensionless, including drag-dropped paths) or accepts an inline title.

Input: $ARGUMENTS (a file path, or a title for interactive creation)

Step 1: Parse Input

Determine the input mode from $ARGUMENTS:

  • File path: if the argument points to an existing file (any path, any extension or none), treat it as a draft to extract from. This covers .claude-work/scratchpads/*.txt, markdown files, extensionless files, and drag-dropped paths from the terminal.
  • Inline title: otherwise, treat the entire argument as the issue title for interactive creation.

Step 2: Extract Issue Content

From File

Read the file and extract:

  • Title: the first #-level heading (strip the # prefix)
  • Body: everything after the title heading

If the file contains a ## Parent or Parent: #NN line, extract the parent issue number for sub-issue linking in Step 6.

Scan the body for dependency relationships to other GitHub issues. The user may express these in natural language — "depends on X", "needs X first", "waiting on X", "blocked by X", "this blocks Y", "unblocks Z", etc. — not with any required phrasing. For each relationship you detect, determine the direction (blocked-by or is-blocking) and extract the full GitHub issue URL. These can reference issues in any repository, including the current one. There can be zero, one, or multiple such relationships.

If the file contains a **Target repo:** owner/repo line (e.g., **Target repo:** couimet/my-claude-skills), extract it as the target repository override. When no target repo line is present, infer owner/repo from the current git remote (gh repo view --json owner,name).

From Inline Title

Use the argument as the title. Prompt the user to provide a body, either inline or by pointing to an existing file.

Step 3: Sanitize Body

Strip references to ephemeral local paths that don't exist on GitHub:

  • .claude-work/ paths (scratchpads, notes, questions, commit-msgs, breadcrumbs)

Print the list of stripped references so the user can verify nothing important was removed. Format as:

Stripped ephemeral references:
- .claude-work/issues/42/scratchpads/0001-plan.txt (line 12)
- .claude-work/issues/42/questions/0001-scope.txt (line 28)

If no ephemeral references were found, print:

No ephemeral references found. Body is clean.

Step 4: Discover Labels and Prompt for Selection

Follow /label-discovery to fetch labels, classify them, and prompt the user. Pass --repo owner/repo when a target repo override was extracted in Step 2.

Step 5: Create the Issue

Use /note with description issue-body to save the sanitized body to a timestamped file in the issue's notes folder (e.g., .claude-work/issues/<ID>/notes/YYYYMMDD-HHMMSS-issue-body.txt). This keeps the body traceable alongside other working files and avoids heredoc compound commands that don't match allowed-tools globs.

Before creating the issue, append a footer line to the body file identifying the skill that generated it. Precede it with a blank line so it stands out from the body content. Read the version: field from this SKILL.md's front matter to fill <VERSION>. Use the Edit tool to append to the body file:


Generated by /create-github-issue v<VERSION> • [my-claude-skills](https://github.com/couimet/my-claude-skills)

Then create the issue with a simple one-liner (pass --repo owner/repo when a target repo override was extracted in Step 2; omit it to use the current git remote):

gh issue create --title "<TITLE>" --label "<LABEL1>,<LABEL2>" --body-file <BODY_FILE_PATH>
gh issue create --repo owner/repo --title "<TITLE>" --label "<LABEL1>,<LABEL2>" --body-file <BODY_FILE_PATH>

Omit the --label flag entirely when no labels are selected. Capture the returned issue URL.

Step 6: Link as Sub-Issue (If Parent Specified)

If a parent issue number was extracted in Step 2, link the new issue as a sub-issue using the link-sub-issue.sh script.

Parse OWNER, REPO, and CHILD_NUMBER from the issue URL returned in Step 5 (https://github.com/{OWNER}/{REPO}/issues/{CHILD_NUMBER}). If a target repo override was extracted in Step 2, use that owner/repo instead.

Run the script once per child issue to link:

~/.claude/skills/create-github-issue/link-sub-issue.sh --owner "$OWNER" --repo "$REPO" --parent "$PARENT_NUMBER" --child "$CHILD_NUMBER"

The script handles all GraphQL calls internally, using jq -n to build payloads via temp files to avoid zsh history expansion stripping ! from GraphQL type annotations (String!, Int!, ID!). It prints linked #<child> → #<parent> on success or an error message on failure (exit 1).

If the script fails, note it in the Step 8 report as:

Sub-issue linking: failed (<error summary>). Link manually if needed.

If no parent was specified, skip this step.

Step 7: Link Dependencies (If Blocked By / Is Blocking Specified)

If Blocked by or Is blocking URLs were extracted in Step 2, link each dependency using the link-dependency.sh script.

Parse OWNER, REPO, and SUBJECT_NUMBER from the issue URL returned in Step 5 (https://github.com/{OWNER}/{REPO}/issues/{SUBJECT_NUMBER}). If a target repo override was extracted in Step 2, use that owner/repo instead.

For each dependency URL detected in Step 2, parse TARGET_OWNER, TARGET_REPO, and TARGET_NUMBER from the URL, then run:

~/.claude/skills/create-github-issue/link-dependency.sh \
  --subject-owner "$OWNER" \
  --subject-repo "$REPO" \
  --subject-number "$SUBJECT_NUMBER" \
  --target-owner "$TARGET_OWNER" \
  --target-repo "$TARGET_REPO" \
  --target-number "$TARGET_NUMBER" \
  --relationship "$RELATIONSHIP"

Where $RELATIONSHIP is blocked-by (when this issue depends on the other) or is-blocking (when this issue unblocks the other), as inferred from the natural language in the body. The script handles all GraphQL calls internally, using jq -n to build payloads via temp files to avoid zsh history expansion stripping ! from GraphQL type annotations (String!, Int!, ID!). It maps the relationship to addBlockedBy argument order and prints blocked #<subject> ← #<target> or blocking #<subject> → #<target> on success.

If the script fails, note it in the Step 8 report as:

Dependency linking: failed for <URL> (<error summary>). Link manually if needed.

To add dependencies to an existing issue (rather than at creation time), use /add-github-dependency <blocked-by|is-blocking> <issue-url>.

If no dependency URLs were extracted, skip this step.

Step 8: Report

Print a summary:

Created: <ISSUE_URL>
Title: <TITLE>
Labels: <LABEL1>, <LABEL2>
Parent: https://github.com/{OWNER}/{REPO}/issues/{PARENT_NUMBER} (linked as sub-issue)  ← only if parent specified AND linking succeeded
Sub-issue linking: failed (<error summary>). Link manually if needed.  ← only if parent specified AND linking failed
Blocked by: <URL> (linked)  ← one per successful blocked-by dependency
Is blocking: <URL> (linked)  ← one per successful is-blocking dependency
Dependency linking: failed for <URL> (<error summary>). Link manually if needed.  ← one per failed dependency

Formatting: see /prose-style for hard-wrap, code-reference, and GitHub-reference rules.

Keep looking

Skills are one crate of 608,666. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.