HExoskeleton

Use when Ground Truth alignment finds memory contamination and the user explicitly requests scoped cleanup with /cleanse.

Use when registering, reviewing, merging, or rebuilding HXSK glossary term definitions after glossary-detect suggests a candidate or the user invokes /define.

Use when staged changes exist requiring qlty checks, logical split detection, or HXSK phase-scoped conventional commit creation.

Use when validating code for circular imports, layer violations, or design doc logic errors before merging architecture-level changes.

Use when .hxsk/.bootstrap-version is missing (fresh install) or exists (verify/update) to initialize HExoskeleton.

Use when shell scripts need linting/formatting before execution or commit to ensure code quality.

Use when analyzing an existing codebase to generate ARCHITECTURE.md and STACK.md, or when mapping project structure, dependencies, patterns, integrations, and technical debt before planning.

Use when debugging fails 3 times, same approach repeats, or context usage exceeds 60% to trigger state dump and recovery.

Use when local changes pass A-E quality checks, lessons-learned review, and are under 500 lines of production code, ready for gh pr create.

Use when debugging bugs to find root causes, or after 3 failed fix attempts to stop and document state.

Use when PLAN.md or SPEC.md exists and tasks require splitting into waves for parallel execution in isolated worktrees.

Use when modifying markdown files, preparing PRs, or suspecting link/structure inconsistencies in documentation.

Use when claiming work is complete, successful, or fixed, and empirical evidence is required to validate the change before proceeding.

Use when a PLAN.md file exists and requires execution to implement tasks atomically with commits.

Use when a session ends, work must pause, or another agent will continue the task to commit changes, run tests, and store handoff memory.

Use when modifying any existing file (excluding new standalone files) to analyze dependencies and calculate impact score.

Use when starting sessions, making architecture decisions, finding root causes, or ending sessions to store or recall project knowledge.

Use when PLAN.md exists after /plan and before /execute to validate 6 dimensions: requirement coverage, task completeness, dependencies, key links, scope sanity, and verification derivation.

Use when SPEC.md exists and you must decompose phase goals into exactly 2-3 tasks with specific file paths, actions, and verification criteria before execution.

Use when a PR number or URL is provided to conduct a comprehensive code review and issue a merge verdict based on blocker severity.

Use when code smells are identified and tests or specs exist to refactor without changing behavior.

Use when proving a skill alters agent behavior under pressure by comparing rule violations without the skill against compliance with it.

Use when code exists but needs validation for stubs, wiring, and anti-patterns to confirm phase completion.

Use when comparing 3-5 solutions to generate a decision report with TCO, status quo risks, and conclusion-first recommendations.

Use when a SKILL.md description, trigger, Quick Reference, or Iron Laws feels vague, too long, or fails to activate the skill correctly