agent-loop-ext
Crash-resilient external agent loop with state persistence and CI/CD integration
Select and execute the canonical refresh, existing-project extraction, or legacy AIWG context regeneration branch
$ npx -y skills add jmagly/aiwg --skill aiwg-regenerate --agent claude-codeHow it fires
How this skill gets triggered: by you, by Claude, or both.
/aiwg-regenerateContext preview
The summary Claude sees to decide when to auto-load this skill.
Select and execute the canonical refresh, existing-project extraction, or legacy AIWG context regeneration branch
namespace: aiwg name: aiwg-regenerate platforms: [all] kernel: true description: Select and execute the canonical refresh, existing-project extraction, or legacy AIWG context regeneration branch script: entrypoint: run.sh runtime: bash cwd: project-root argsHint: '[--workspace|--existing-project|--full-inject] [--provider <name>] [--dry-run|--apply] [--force]'
The CLI is the deterministic source of truth for context regeneration. An unqualified invocation previews transactional adoption for an established, not-yet-extracted repository; fresh or already-adopted projects route to the canonical refresh. Mutation of an inferred adoption always requires an explicit `--apply`. Do not make users choose a branch unless they want to override the selection.
Select and load exactly one linked branch before execution:
refresh for new and already-migrated projects.
transactional adoption of stable project metadata and provider context into the canonical graph; preview is the default and `--apply` is required.
branch; embeds normalized AIWG context inside the provider startup file.
# Intelligent default (recommended user surface) aiwg run skill aiwg-regenerate # Canonical default aiwg regenerate --workspace [--provider <name>] [--dry-run] [--force] # Existing-project transactional adoption aiwg regenerate --existing-project [--provider <name>] [--dry-run|--apply] # Legacy compatibility aiwg regenerate --full-inject [--provider <name>] [--dry-run]
`--legacy` aliases `--full-inject`. The CLI rejects conflicting branches, unknown options, and missing provider values with usage status.
`--existing-project` rejects `--force` and all `--no-*-md` partial-write flags. It refuses possible credentials and ambiguous directive conflicts, previews the exact extracted block and transaction targets, and prints the rollback command after apply.
Explicit branch flags always win. Without one, the executable selector checks the canonical extractor's complete stable-source inventory and managed markers. A first-time established project is previewed; rerun with `--apply` to adopt it. An already-extracted or genuinely signal-free project uses `--workspace`.
Supported shared controls are `--dry-run`, `--provider <name>`, `--force`, `--no-aiwg-md`, `--no-agents-md`, and `--no-workspace-md`. The last option is implicit in legacy mode.
1. For an established project, preview the selected branch first. 2. Preserve operator-authored content outside managed regions. 3. Never copy possible credentials into generated context. 4. Execute with `aiwg run skill aiwg-regenerate -- <flags>` when the provider supports executable skills, or run the equivalent `aiwg regenerate` command. 5. Report the selected branch, provider, changed targets, backups, warnings, validation result, and rollback path.
Use `aiwg refresh` instead when frameworks, agents, skills, rules, commands, or provider deployments also need to be updated.
Reusable project context and specialist workflows for the AI tools you already use. Plan software, coordinate specialist reviews, prepare campaigns, investigate incidents, organize research, curate media, and maintain operational knowledge.
Repo: jmagly/aiwg
Crash-resilient external agent loop with state persistence and CI/CD integration
Detect requests for iterative autonomous agent loops and route to the appropriate loop executor
Automatically execute tests when code-generating agents modify source files, enforcing the execute-before-return pattern
Enable agent loops to learn from similar past tasks and share patterns across loops
Query and manage the executable feedback debug memory
Execute tests on generated code and iterate until passing