Irinṣẹ́
Select, ready, and use a companion tool to answer the current engineering question.
1. Bound the need and select the tool
Identify the question the tool should answer, exact candidate, required evidence, acceptable fallback, mutation authority, and freshness boundary. HEAD alone is not an exact identity for analyzed uncommitted content. Prefer direct source reading, project-native commands/wrappers, and existing safe tools when they answer the question proportionately.
Select the smallest useful tool. Do not recommend every missing tool or route ordinary shell/search work through irinse. Do not introduce a new tool merely to make one proof step convenient when current project/platform capability already provides sufficient evidence. For a consuming task that did not already authorize tooling expansion, adopting a new dependency/tool/integration is a scope-expansion fact to return to that owner rather than silently changing its surface.
Load only the selected reference:
- tldr-code for compact structural, flow, impact, quality, or security leads;
- ast-grep for syntax-aware search or structural rewriting;
- Semgrep for repeatable bug, security, or architecture rules;
- IntelliJ MCP for IDE-backed navigation, code insight, run configurations, and debugging; or
- Firecrawl for agent-oriented search, scrape, crawl, or browser-backed web acquisition when ordinary fetch/search cannot reliably expose the needed evidence.
For another named tool, use current official documentation. Stable references describe capabilities and boundaries; volatile commands/version support belong to current tool/project evidence.
If the request needs only selection, return the chosen tool, distinct benefit, bounded question, safe fallback, and readiness uncertainty without loading setup procedure.
2. Operate when needed
When the selected tool must be readied, installed, configured, authenticated, integrated, used, upgraded, removed, or rolled back, read operate. Do not mutate beyond explicit authority.
3. Return the evidence
Return the bounded need/candidate, tool identity/version, readiness state, authorized operation, observations with provenance, corroboration, configured rule/threshold/severity when the tool reports one, coverage limits/errors, freshness, and any consuming-task scope-expansion fact. Do not turn tool output or scores into architecture, implementation, security, maintainability, or review verdicts; return the evidence to the owning workflow.
1---2name: irinse3description: Select, install, configure, integrate, use, verify, upgrade, or remove one companion engineering tool. Use when a coding workflow needs bounded evidence from tldr-code, ast-grep, Semgrep, IntelliJ MCP, Firecrawl, or another named tool, or when a useful required tool is missing; exclude architecture, implementation, security, maintainability, and review verdicts.4---56# Irinṣẹ́78Select, ready, and use a companion tool to answer the current engineering question.910## 1. Bound the need and select the tool1112Identify the question the tool should answer, exact candidate, required evidence, acceptable fallback, mutation authority, and freshness boundary. HEAD alone is not an exact identity for analyzed uncommitted content. Prefer direct source reading, project-native commands/wrappers, and existing safe tools when they answer the question proportionately.1314Select the smallest useful tool. Do not recommend every missing tool or route ordinary shell/search work through `irinse`. Do not introduce a new tool merely to make one proof step convenient when current project/platform capability already provides sufficient evidence. For a consuming task that did not already authorize tooling expansion, adopting a new dependency/tool/integration is a scope-expansion fact to return to that owner rather than silently changing its surface.1516Load only the selected reference:1718- [tldr-code](references/tldr-code.md) for compact structural, flow, impact, quality, or security leads;19- [ast-grep](references/ast-grep.md) for syntax-aware search or structural rewriting;20- [Semgrep](references/semgrep.md) for repeatable bug, security, or architecture rules;21- [IntelliJ MCP](references/intellij-mcp.md) for IDE-backed navigation, code insight, run configurations, and debugging; or22- [Firecrawl](references/firecrawl.md) for agent-oriented search, scrape, crawl, or browser-backed web acquisition when ordinary fetch/search cannot reliably expose the needed evidence.2324For another named tool, use current official documentation. Stable references describe capabilities and boundaries; volatile commands/version support belong to current tool/project evidence.2526If the request needs only selection, return the chosen tool, distinct benefit, bounded question, safe fallback, and readiness uncertainty without loading setup procedure.2728## 2. Operate when needed2930When the selected tool must be readied, installed, configured, authenticated, integrated, used, upgraded, removed, or rolled back, read [operate](references/operate.md). Do not mutate beyond explicit authority.3132## 3. Return the evidence3334Return the bounded need/candidate, tool identity/version, readiness state, authorized operation, observations with provenance, corroboration, configured rule/threshold/severity when the tool reports one, coverage limits/errors, freshness, and any consuming-task scope-expansion fact. Do not turn tool output or scores into architecture, implementation, security, maintainability, or review verdicts; return the evidence to the owning workflow.
quantipixels