Kickoff
Claude Code plugin: tech-lead interview -> multi-file execution plan -> sprint-by-sprint execution -> zero-trust audit -> handoff. Guarded against both lazy shortcuts and over-engineering.
npx -y skills add Zhen-Bo/request-devflow-plan --skill kickoffAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
2 things to look at
- 27 days oldThe repository was created 27 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
- 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Start or continue an execution session for a DevFlow plan. The span argument sets the scope — one sprint, a sprint range (batch), or a whole phase. Collects the maintainer's review sign-off interactively when the previous unit is unsigned; continue mode resumes from the ledgers, reconciles HEAD, and requires a green start.
SKILL.md
5.3 KB, as published. Nobody here has run it
request-devflow-plan:kickoff
Start or resume execution of a DevFlow plan. $ARGUMENTS: a target unit (sprint N, sprint N-M, or phase N) and optionally continue — in either order (kickoff continue sprint 1 and kickoff sprint 1 continue are equivalent).
Flow
Step 1 — Locate the plan folder
Scan the target repo's docs/*/ for a DevFlow plan: a README containing a doc map with a 00-iron-rules.md beside it. Confirm with the user; list options if several match. Only one plan executes actively in a repo at a time — starting a second requires the first to be finished, or paused at a commit boundary with its session record written.
Step 2 — Build plan memory
Read: README.md (doc map, global gates, cross-repo dependencies), PROGRESS.md (done vs pending), implementation-notes.md (latest session records + deviations), and the target unit's sprint doc(s).
Step 3 — Branch
Fresh start (no continue, no existing progress on the unit) → template A semantics
- Check the previous unit's maintainer sign-off in PROGRESS.
- Unsigned → collect the review results interactively, one question at a time: (1) quiz passed? — the quiz grades itself in the maintainer's browser and records nothing by design; their answer here IS the record; (2) one reply per open deviation, offering the maintainer guide's safe default for each; (3) any adjustment instructions. Write the answers into PROGRESS (sign-off row) and the Deviations status column yourself — the maintainer never needs to edit ledgers by hand (hand-filling first is allowed; then this step sees the sign-off and skips the questions). Handle adjustments first, as normal commits with gates. (This is the same content template B's review-results section carries in the copy-paste flow.)
- Signed → proceed.
- Start per template A: full plan read-through, baseline gates, entry conditions, publish the HTML implementation plan, then TDD.
Resume (continue) → template B semantics (resume flavor)
- Read the last session record's "next step" in implementation-notes §3.
- Batch protocol: reconcile the ledgers against HEAD.
- Green start: all gates green before touching anything.
- Confirm entry conditions; publish or refresh the HTML implementation plan, then continue TDD.
Post-audit fixes never run as loose adjustments: a FAIL audit becomes a remediation unit (plan skill, remediation mode), which then starts here as a normal fresh unit.
Step 4 — Span
sprint N(default): sprint by sprint; stop at sprint end and wait for review (template A stop semantics).sprint N-M(batch): run the range; produce each sprint's report and verification without stopping. Deviations accumulate in the ledger under conservative continuation — there is deliberately no batch safety valve.phase N: batch the whole phase; the phase boundary is a mandatory stop.
First line of the session output: plan path, target unit, span.
Next steps block (every stop ends with one)
Print full, copy-paste-ready command lines — always the complete /request-devflow-plan:<command> <args> form with the actual unit numbers substituted. Shorthand and unfilled placeholders in the printed output are violations. Recommended command first; alternatives after, each with one line on when to choose it. Every follow-up command here runs in a fresh context — print the /clear advisory line immediately before the command lines. Example of a correct block after finishing sprint 2 of a 5-sprint plan:
Next steps — after you review the report, pass the quiz, and answer the open deviations:
Run /clear (or open a new session) to empty the context, then:
▶ /request-devflow-plan:kickoff sprint 3 — next unit (your sign-off is collected there)
/request-devflow-plan:kickoff sprint 3-4 — batch the remaining units, one review at the end
Per state:
- Unit complete, feature units remain → the block above, with the real next unit number.
- Unit complete and it was the last feature unit → recommend
/request-devflow-plan:audit, stating explicitly: run it in a NEW session — the plan's acceptance unit is next. - Stopped mid-unit (context exhausted, interruption) → recommend
/request-devflow-plan:kickoff continue sprint <actual current unit number>.
Design principles
- The plan folder is the single source of truth; this skill launches it and copies no rules. When an older plan folder and newer skill semantics disagree, the plan folder wins; ledgers it predates (e.g.
audit-log.md) are created on first use. - The plan's
kickoff-prompts.mdis the portable fallback — everything works by copy-paste without this plugin. - Sign-off collection is interactive and never assumes success.
Never
- Modify plan documents (or repair the executor's acceptance evidence)
- Launch past an unsigned previous unit without collecting the review
- Start on red gates