hebstr

@hebstr source repo

10 skills
0 followers
6 hours ago last updated

▌
Sweep · hebstr bundle
User-invocable ONLY via `/audit:sweep`. Does not auto-trigger on mentions of "full review", "audit complet", "review complète", "review this project", "audit this repo", "thorough review", "diagnostic complet", "état des lieux du projet", or any phrasing requesting a multi-angle project assessment. Full project review: detects project type and size, spawns specialist background agents with disjoint scopes (architecture, quality, tests, docs), consolidates findings into one deduplicated report sorted by severity, then offers an interactive walkthrough. Not for: single-file reviews, PR or diff reviews, or non-code document reviews (papers, resumes, CVs).

0
installs
▌
Reco · hebstr
Deep recommendation backed by external sources (official documentation + community practice + verified citations). For an unsourced recommendation, ask the model directly; use this skill when external grounding is what makes the answer load-bearing. Without arguments: ask the user what they want a recommendation on. With arguments: research the topic via parallel agents (official docs via WebFetch, community practice via WebSearch), then synthesize a structured recommendation. User-invocable ONLY via /workflow:reco; does not auto-trigger on mentions of 'recommendation', 'reco', 'recommend', 'suggest', 'advice', or French equivalents ('reco', 'recommande', 'recommandation', 'conseille', 'suggère').

0
installs
▌
Sync · hebstr
Scan all files in the current directory and subdirectories, identify files that are stale relative to recent changes, and update them. Always performs a cross-repo semantic consistency scan with parallel agents. User-invocable ONLY via /workflow:sync; does not auto-trigger on mentions of 'sync', 'synchronize', 'synchronization', 'stale files', 'stale data', 'consistency check', 'cross-repo check', or French equivalents ('synchroniser', 'synchronisation', 'cohérence', 'fichiers obsolètes').

0
installs
▌
Write · hebstr bundle
Invoke ONLY when the user explicitly types `/workflow:write`. Do not auto-trigger on mentions of 'draft', 'edit', 'proofread', 'polish', 'rewrite', 'sound natural', or French equivalents ('écris', 'rédige', 'relis', 'corrige', 'polis', 'dégraisse', 'retravaille'). Strips AI writing patterns and rewrites prose to sound natural in English or French. Not for code comments, commit messages, or inline docs.

0
installs
▌
Blindspot · hebstr bundle
User-invocable ONLY via `/audit:blindspot`. Does not auto-trigger on mentions of "circular review", "review circulaire", "self-review check", "second opinion", "external judge", or any phrasing requesting circularity countermeasures. Runtime enforcement via `disable-model-invocation: true`: the Skill tool cannot launch this skill; user invocation is mandatory. Reviewer orchestrator: detects when an audit skill (skill-adversary, mcp-adversary, sweep, critical-code-reviewer) is about to review an artifact that shares its codebase, prompts, or model family, then runs cross-model judging via OpenRouter, followed by convergence analysis over the resulting verdicts. Not for: general code review, PR review, plain skill-adversary/mcp-adversary/sweep invocations without circularity concern, or LLM evaluation methodology discussions.

0
installs
▌
Walkthrough · hebstr bundle
User-invocable ONLY via `/audit:walkthrough`. Does not auto-trigger on mentions of "review", "review findings", "review report", "walk through", "walkthrough", "go through one by one", "triage", "DEFERRED.md", "revisit deferred", or French equivalents ("revue", "passer en revue", "trier", "reprendre les findings", "revoir les éléments différés"). Interactive, point-by-point walkthrough of a review report produced by any Claude Code review skill (skill-adversary, critical-code-reviewer, or any other). Three modes: **orchestrator mode** (provide a target + optional `--reviewer` flag: detects deployment context, calibrates severity, launches the reviewer, then walks through its report), **walkthrough-only mode** (processes an existing report from the conversation), and **revisit-deferred mode** (processes the project's `DEFERRED.md` backlog as the input source). Parses review findings and processes each one at a time: re-evaluates validity, proposes and applies fixes, checks impacted files for regressions, and w

0
installs
▌
Continue · hebstr
Flush durable facts to memory, update PLAN.md, and print a minimal continuation prompt. No handoff document is written. User-invocable ONLY via /workflow:continue; does not auto-trigger on mentions of 'continue', 'handoff', 'session handoff', 'continuation prompt', 'save state', 'persist', 'resume', or French equivalents ('continuer', 'passation', 'reprise', 'sauvegarder l''état').

0
installs
▌
MCP Adversary · hebstr bundle
User-invocable ONLY via `/audit:mcp-adversary`. Does not auto-trigger on mentions of "review my MCP", "audit MCP server", "tool descriptions", "schema review", "tool quality", or French equivalents ("auditer mon MCP", "passer mon MCP au crible"). Adversarial reviewer for MCP servers: reads tool descriptions, parameter schemas, and implementation code, then reports tool-selection ambiguity, discoverability gaps, schema anti-patterns, semantic drift between description and behavior, error-handling inconsistencies, and undocumented workflow dependencies. Not for: general code review (use posit-dev:critical-code-reviewer), security scanning (use mcp-scan), reviewing Claude Code skills (use skill-adversary), creating MCP servers, or non-MCP tool/API review.

0
installs
▌
Skill Adversary · hebstr bundle
User-invocable ONLY via `/audit:skill-adversary`. Does not auto-trigger on mentions of "audit this skill", "review SKILL.md", "find flaws", "adversary review", "attack this skill", "trigger edge cases", "stress-test", or French equivalents ("auditer cette skill", "trouver les failles", "passer ce skill au crible"). Adversarial reviewer for Claude Code skills: reads a skill's full directory (SKILL.md, agents, docs, templates) and reports trigger edge cases (false positives and negatives), instruction ambiguities, contradictions, cross-file coherence issues, and gaps. Not for: general code review (use posit-dev:critical-code-reviewer), reviewing non-skill files, or creating/editing skills (use skill-creator).

0
installs
▌
Doc Structure · hebstr
Audit and reorganize project documentation layout (CLAUDE.md for Claude rules, README.md for human docs). User-invocable ONLY via `/workflow:doc-structure [<project-path>]`; the path is optional and defaults to the current working directory. For monorepos, invoke once per sub-package path. Does not auto-trigger on mentions of CLAUDE.md, README.md, documentation, or sync. Not for code-level docstring/README drift, splitting one doc file into a `docs/` tree, or reorganizing sections within a single CLAUDE.md or README.md.

0
installs

10 published skills