SuperAPI AI Directory

execute-issue

General & other

View repository

Updated 10-15 17:32

About this skill

Execute Issue

Implement one specific issue from start to finish. This skill is intentionally independent from Ralph: it does not select from an AFK queue, run repeated iterations, or assume a wrapper script. It executes the issue the user gives you or the issue already present in context.

Inputs

Accept any of these inputs:

  • A local issue path, such as docs/issues/<change>/<issue>.md
  • A folder of related local issue files, such as docs/issues/<change>/
  • Pasted issue text
  • A GitHub, Linear, or tracker issue already available in context
  • A bug or task described in the current conversation

If the target issue is ambiguous, ask one concise question to identify it. If the issue is clear, proceed without asking.

Process

1. Understand the issue

Read the issue and extract:

  • The problem to solve
  • Acceptance criteria
  • Explicit non-goals
  • Blockers, dependencies, or HITL requirements
  • Expected tests or verification steps

If the issue is marked blocked or requires human input, stop and ask before implementing unless the user explicitly provided the missing decision.

2. Inspect current state

Explore the repo enough to verify the issue against reality:

  • Existing behavior and failing tests, if any
  • Nearby implementation patterns
  • Public interfaces and boundaries affected by the change
  • Existing tests that describe the behavior
  • Project-specific commands in README, package scripts, CI config, Makefile, task files, or solution files

Keep exploration proportional to the issue. Do not refactor unrelated code.

3. Check current external documentation

For third-party frameworks, libraries, SDKs, APIs, CLI tools, cloud services, endpoints, configuration options, or syntax, fetch current documentation using the repository's required documentation lookup flow.

In repositories that require ctx7, use:

  1. npx ctx7@latest library <name> "<specific question>"
  2. npx ctx7@latest docs <libraryId> "<specific question>"

Do not use external documentation lookup for general programming concepts or project-local business logic.

4. Implement with focused tests

Prefer test-first work when practical:

  • Reproduce the issue with a failing test, or identify an existing failing test
  • Implement the smallest coherent change that satisfies the issue
  • Keep tests at public boundaries where possible
  • Avoid implementation-detail assertions unless the project already uses them for this layer

If the issue is too small for a new test, state why in the final response.

5. Verify

Run the most relevant feedback loops before finishing:

  • Targeted tests for the changed behavior
  • Broader test suite if the change touches shared behavior
  • Build, typecheck, lint, or format commands expected by the repo

If a command cannot run because of missing services, dependencies, credentials, or sandbox restrictions, record the blocker clearly.

6. Update the issue record

If the issue came from a local Markdown file:

  • Move it to done/ only when acceptance criteria are met and verification has run or the blocker is explicitly acceptable.
  • If incomplete, append a short progress note with completed work, verification status, and remaining blockers.

If the issue came from a tracker and tracker tools are available, update the tracker only if the user asked you to. Otherwise, summarize what should be posted.

7. Commit when appropriate

Commit only when the user requested commits or the issue workflow clearly expects commits.

When committing, include:

  • What changed
  • Key decisions
  • Tests or verification run
  • Remaining blockers, if any

Do not include unrelated untracked files.

Final Response

Report:

  • What was implemented
  • Files changed
  • Verification run and results
  • Issue update or remaining blocker

Keep the response concise. Do not paste large diffs.