From xlfg-engineering
Autonomous xlfg SDLC run. Batches hidden recall, intent, context, plan, implement, verify, review, and compound skills end-to-end.
npx claudepluginhub flrngel/xlfg --plugin xlfg-engineering[feature, bugfix, investigation, or delivery request]Use this when the user wants a serious engineering run with PM, UX, Engineering, QA, and release discipline. INPUT: `$ARGUMENTS` Treat this invocation as **one autonomous run**. `/xlfg` is the conductor for a **batch of hidden phase skills**. Load only the current phase skill, execute it, then move to the next one. Do not inline the whole workflow from memory and do not ask the user to run internal skills or phase commands. ## Run contract - divide specialist work into atomic task packets: one clear mission in, one required artifact out - keep `spec.md` as the single source of truth fo...
Share bugs, ideas, or general feedback.
Use this when the user wants a serious engineering run with PM, UX, Engineering, QA, and release discipline.
INPUT: $ARGUMENTS
Treat this invocation as one autonomous run.
/xlfg is the conductor for a batch of hidden phase skills. Load only the current phase skill, execute it, then move to the next one. Do not inline the whole workflow from memory and do not ask the user to run internal skills or phase commands.
spec.md as the single source of truth for intent, chosen solution, task map, proof status, and PM / UX / Engineering / QA / Release notesspec.mdxlfg helper CLI when availableO1, O2, ...) before broad repo fan-outxlfg init (or xlfg prepare) when the helper exists; otherwise perform the equivalent bootstrap manually.RUN_ID with xlfg start "$ARGUMENTS" when the helper exists; otherwise create the lean core run manually.DOCS_RUN_DIR=docs/xlfg/runs/<RUN_ID> and DX_RUN_DIR=.xlfg/runs/<RUN_ID>.After startup, write .xlfg/phase-state.json with this initial state:
{
"run_id": "<RUN_ID>",
"phases": ["recall","intent","context","plan","implement","verify","review","compound"],
"completed": [],
"loopback_count": 0,
"max_loopbacks": 2,
"block_count": 0
}
After each phase skill returns successfully, add that phase name to completed and reset block_count to 0. Write the file back immediately. A Stop hook reads this file to prevent the conductor from ending before all phases are done.
Invoke these hidden skills in this exact order, always passing RUN_ID:
xlfg-engineering:xlfg-recall-phasexlfg-engineering:xlfg-intent-phasexlfg-engineering:xlfg-context-phasexlfg-engineering:xlfg-plan-phasexlfg-engineering:xlfg-implement-phasexlfg-engineering:xlfg-verify-phasexlfg-engineering:xlfg-review-phasexlfg-engineering:xlfg-compound-phaseUse the Skill tool to load each phase just-in-time instead of carrying all phase instructions in the entrypoint.
Before dispatching any xlfg specialist, preseed the required artifact yourself and use a packet that starts with these machine-readable lines:
PRIMARY_ARTIFACT: <exact path>
FILE_SCOPE: <bounded files or paths>
DONE_CHECK: <single honest check or NONE>
RETURN_CONTRACT: DONE|BLOCKED|FAILED <artifact-path> only
PRIMARY_ARTIFACT with Status: IN_PROGRESS, the mission, and a short remaining checklist before the specialist starts broad reading.PRIMARY_ARTIFACT and explicit RETURN_CONTRACT; file-backed artifact progress is the only accepted basis for waiting.Status: DONE or Status: BLOCKED or Status: FAILED, and contains concrete edits, findings, checks, logs, or cited facts.SendMessage using its returned agent ID so it continues from prior state instead of starting over. If no agent ID is available or resume is unavailable, re-dispatch the exact same packet once.xlfg-engineering:xlfg-intent-phase has written the intent contract and objective groups in spec.md.resolution: needs-user-answer, stop and ask at most three concise numbered blocking questions. Do not continue to context, planning, or coding until the answer arrives.test-readiness.md is not READY after planning, return to xlfg-engineering:xlfg-context-phase and xlfg-engineering:xlfg-plan-phase yourself until the plan is repaired or a true human-only blocker is explicit.xlfg-engineering:xlfg-implement-phase, then rerun xlfg-engineering:xlfg-verify-phase. Increment loopback_count in .xlfg/phase-state.json each time. Max 2 loopbacks — after 2 verify-fix cycles, stop and escalate to the user with a summary of what failed and why.xlfg-engineering:xlfg-implement-phase, then rerun verify and review. This also counts toward the loopback limit.Finish with a concise status summary that includes RUN_ID, what changed, proof status, residual risk, objective completion status, and follow-ups if any.