Use cases¶
Repo Maintainer executes a focused task against a repository. The task can come from a scheduled configuration, a script, or an authenticated webhook integration. Each route uses the same editing boundaries, independent checks and result.
Prism: reference-driven maintenance¶
Prism contains independent Rust SDKs for OpenAI, Gemini and Groq, plus shared transport code.
Trigger: weekly GitHub Actions schedule, with a manual dry run.
Context: official provider changelogs and API references in sources, plus the repository's implementation and contract tests.
Work: identify a documented compatibility gap, implement a focused fix, add offline contract tests, and update examples where needed.
Output: a checked maintenance PR, an unchanged result when no work is needed, or a blocker with diagnostics. Full CI and human review precede merge.
Prism template and setup prompt · Copyable YAML · Web context
The template illustrates Prism's use case. It does not imply that every provider has been live-tested or that a particular consumer rollout is complete.
Vulnerability reports to Slack¶
Trigger: your scanner sends a signed webhook to your gateway. The gateway authenticates it, deduplicates it, and sends a vulnerability-report dispatch to a fixed GitHub repository.
Context: a bounded list of findings: ID, severity, title and details.
Work: validate the payload and format a concise findings list. This reporting path is deterministic; it spends no model calls and labels findings as scanner claims rather than verified vulnerabilities.
Output: a report in your configured Slack channel. Reporting does not automatically start remediation.
Security setup prompt · Example report event
Slack delivery uses an incoming webhook bound to the selected channel. Slack’s incoming-webhook documentation.
Vulnerability fixes as pull requests¶
Trigger: an authenticated vulnerability-fix dispatch for one finding.
Context: the finding's reproduction details, repository source and tests, plus trusted vendor references configured in code.
Work: investigate applicability, establish the vulnerable path, add a regression test, and repair the root cause. The agent should explain why the previous behavior was vulnerable and how the changed behavior addresses it.
For example, a cross-tenant document lookup might require adding an ownership constraint and testing that tenant B cannot retrieve tenant A's document. Updating dependencies would not repair that application logic. The template's sample is fictional and exists to demonstrate that distinction.
Output: an independently checked patch published as a GitHub PR. Slack gets the run outcome and a link to the Actions run containing the PR and evidence. An inapplicable finding returns unchanged with reasoning; an uncertain or unreproducible finding returns blocked. Passing configured checks is evidence, not a guarantee that every security property has been covered.
Security workflow · Trusted configuration · Example fix event
The workflow separates the coding-agent job from GitHub write access and Slack delivery. No fix is automatically merged. Configure meaningful regression tests and review the patch.
Maintenance from a script¶
Trigger: a local command, background queue, or scheduled CI job.
Context: your Python or TypeScript code provides a configuration object and optional task text. No maintainer YAML file is required.
Work: a bounded bug fix, a small feature, documentation upkeep, test coverage, or a focused refactor within the configured paths.
Output: changed, unchanged or blocked, with artifacts available to the caller. The SDK does not add a scheduler or publish automatically.
Script template and setup prompt · Python SDK · TypeScript SDK
Starting configurations¶
Node, Python, Go, and documentation configurations show language-specific checks. For any task, start with one outcome, narrow file permissions and a test that establishes the intended behavior.