What is Skillkeel Starter?
Skillkeel Starter is a free, MIT-licensed plugin for people who use Claude Code on software projects. It combines hooks that block selected risky shell commands and file writes with eight skills for recurring repository tasks, including changelogs, dependency audits and test-gap reviews. The skills are tested against fixture repositories, and their eval cases and transcripts are included so teams can inspect how they behave before installing them.
Skillkeel Starter Features:
Shell Command Guard: The guard-bash hook blocks selected destructive commands before they run, including hard resets, force-pushes without a lease, recursive deletes, database wipes and Terraform destroys.
Credential Write Checks: The secret-scan hook refuses file writes containing recognized API keys, private keys, JWTs, hard-coded passwords or tokens, and connection URLs with an inline password.
Session Project Brief: At the start of a Claude Code session, session-start supplies a brief with the current Git state and the names of the available skills.
Repository Skills: Eight skills handle tasks such as drafting commit messages and PR descriptions, refreshing README commands and links, auditing dependencies, and finding test gaps.
Skill Eval Records: Each skill includes a fixture repository, expected results and an unedited transcript of a Claude Code run, with eval cases and pass results kept in the repository.
Local Hook Runs: The hooks run on the user's machine and make no network calls; a refusal explains the rule and asks Claude to check with the user.
Skillkeel Starter Benefits:
Fewer Risky Runs: Blocking selected dangerous commands before execution can help prevent accidental damage to project files, databases or shared Git branches.
Earlier Secret Checks: Catching credentials in a proposed write gives developers a chance to remove them before they enter a file.
Less Routine Drafting: Automating repeatable repository chores leaves developers with less manual writing and checking to do for common project updates.
Reviewable Behavior: Fixture-based evals and readable transcripts let a team inspect skill behavior before adopting it in its own repository.
Clearer Decisions: Specific refusal messages explain why an action was stopped, helping users decide whether to revise it or approve it themselves.
Use Cases:
Reviewing Shell Actions: Use the hooks when Claude Code is asked to clean a working tree, change Git settings, reset a repository or run a database operation.
Checking Proposed Writes: Run secret-scan while editing configuration or other project files that could accidentally include credentials.
Preparing Pull Requests: Use the PR-description skill to draft a body from a branch diff, including a test plan based on commands present in the repository.
Maintaining Project Docs: Use changelog and README-refresh when preparing a release or checking whether documentation commands and links still match the project.
Finding Test Gaps: Use test-gap-finder to identify untested modules by recent changes and get specific test-case proposals.