- Home
- Skills
- Agents & Orchestration
- Phase-Gated Implementation
Works with the AI Agents you already use
Phase-Gated Implementation
Keeps your agent inside one assigned phase of a multi-phase plan - verified by typecheck/lint/test, reported honestly, then stopped - so a planner or reviewer never has to re-check work that silently ran ahead.
$2
Phase-Gated Implementation
Example session with this skill installed
I'm the reviewer on this project. Here's the spec: docs/spec/billing-v2.md, and the phase plan is in docs/PLAN.md. Please implement Phase 4: Invoice PDF export only. Stop when it's done and report back.
- Read your context and instructions
- Compiled the phase-gated implementation
- Generated the data export
Phase 4: Invoice PDF export - Report
Re-read: docs/spec/billing-v2.md §4.2 "PDF export" and §4.5
"File naming conventions" before starting, since Phase 3 touched
adjacent invoice code.
Files changed
- src/billing/exportInvoicePdf.ts (new) - renders an Invoice into a
PDF buffer using the existingpdf-libdependency already used
elsewhere in the codebase - src/billing/routes/invoices.ts - added GET
/invoices/:id/export.pdf, reusing the existing auth middleware - src/billing/exportInvoicePdf.test.ts (new) - covers: a standard
invoice, an invoice with zero line items, and a non-existent
invoice ID (expects 404)
Verification
- npm run typecheck -> pass
- npm run lint -> pass (0 warnings)
- npm test -> pass, 14/14 (3 new, 11 existing, all green)
- Manually hit the new endpoint locally against a seeded invoice
and confirmed the downloaded PDF opens and matches the on-screen
invoice totals
Assumptions
- Spec §4.2 doesn't state a filename pattern for the download, so
I usedinvoice-{id}.pdfto match the naming already used by
the CSV export in Phase 2. Flagging this in case you want a
different convention.
Left out on purpose
- §4.5 mentions an optional company-logo watermark on the PDF.
That's listed under Phase 6 in PLAN.md, so I did not implement
it here - noting it so it isn't mistaken for a gap in this phase.
Stopping here. Phase 5 (bulk export) not started.
export.pdf
CSV · data export
Example file from a real run - the skill writes it into your workspace.
Connects securely to your tools. The creator never sees your data.
About this skill
Most agents fail a planner/reviewer workflow the same way: they do more than they were asked, or they say "done" before anything was actually checked.
Phase-Gated Implementation fixes both. It turns your agent into a disciplined implementer that takes exactly one phase of a plan, does only that phase, verifies it with real commands instead of self-assessment, and reports back with full honesty - including what it didn't do.
Built for split workflows. If you run a setup where one agent (or person) owns the plan and spec, and a separate implementer executes phase by phase, this skill is the contract that keeps the implementer trustworthy across dozens of phases without you re-verifying every single one.
What it enforces
One phase, no more. The agent stops the moment its assigned phase is done - even with time and context left. No silent scope creep into "while I'm here" territory.
Spec as source of truth. The full spec is read once up front; the relevant sections are re-read before every phase, because plans shift.
Reference material stays read-only. Screenshots, prior art, and design references inform the work - they never get edited, and they never override an explicit spec requirement.
Green means green. Typecheck, lint, and test all have to pass before a phase is reported done. A lint warning is treated as a failure, not noise.
Fixed boundaries. The plan, the spec, your own instruction files, and git history are off-limits unless you explicitly say otherwise. No committing, no amending, no "helpfully" editing the plan to match the code.
Honest reporting. Every phase ends with: files touched, what was verified and how, and anything left out or assumed - stated plainly, not buried or omitted.
Who this is for
Teams or solo builders running a planner/reviewer split, where one agent plans and a separate implementer (or implementer agent) executes
Anyone managing multi-phase specs or PRDs where completion must be proven by commands, not vibes
Reviewers who are tired of re-auditing implementer work because the implementer quietly did three phases instead of one, or claimed success without running the test suite
What it is not for: solo work with no external plan, one-off single-file fixes, or early exploratory spikes where scope isn't fixed yet - using it there just adds friction.
Drop this skill into any agent that executes phased work, and every phase comes back the same way: scoped, verified, and reported straight.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Agents & Orchestration.
Requires a pre-defined plan and spec document. Does not handle git commits, merges, or plan file edits. Not suitable for exploratory spikes without fixed scope.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 20 days ago
Permissions
Scope isn't defined by file type or directory - it's defined by whichever phase of the plan you assign. The skill only touches files that the current phase's spec sections call for, and treats the plan file, spec file, and git history as read-only unless explicitly told otherwise.