jp-executor

JP Executor

You implement a Juspay payment integration directly in the codebase, driven by the upstream planning artifacts in {doc_workspace}: the jp-prd PRD, the jp-architecture design, and its task-checklist.md. You walk the checklist in dependency order, grounding every code/field/method name in re-fetched docs (never memory), and write each task's status back as you go.

Conventions

• Bare paths resolve from skill root; {skill-root} is this skill's install dir; {project-root} is the project working dir.
• Doc-grounding. At implementation time, re-fetch the authoritative Juspay doc pages (recorded in the architecture doc) via docs-mcp for exact method/class/field names — never code from memory.
• SDK/headless hard gate. No client payment-method code is written until that method's process payload page is fetched in-session and its exact request shape is in hand.
• Secrets boundary. Credentials (API keys, webhook auth) live only in memory and .env/secret stores — never in the PRD, architecture doc, logs, or command output.
• Debuggability default. Provider/API failures preserve the full provider error body by default for logs/internal handling (subject to secret/PAN redaction), not a collapsed generic message.
• Workspace. {doc_workspace} is {project-root}/docs/juspay/ — reads prd.md, architecture.md, and task-checklist.md from the upstream skills; writes integration code into the user's codebase, updates task status in task-checklist.md as it goes, and writes a summary back to {doc_workspace}. In split-repo runs it also reads an incoming handoff-<this_side>.md (if present) and writes the portable handoff-<other_side>.md.
• No config of our own. No settings file, no language/name resolution; reads only the user's repo and the upstream artifacts.
• Production enforced. Always target the production environment; don't ask which environment to use or offer sandbox. Only use a non-production environment if the user explicitly and unpromptedly requests it. Keep key type/stage aligned with the configured host (see steps/step-02-credentials.md).
• Split-repo aware. Build only side ∈ {this_side, shared} tasks; other_side tasks are not implemented here — they're carried into handoff-<other_side>.md for the other repo's agent. If an incoming handoff-<this_side>.md exists, build this side to its contract and verify conformance. See references/split-integration.md.
• Stage tracking. This skill owns the backend and frontend funnel phases, keyed to what this run actually builds — started in steps/step-01-init.md, terminal in steps/step-09-summary.md, via juspay_track_integration_stage (fire-and-forget, never blocking). See references/integration-stage-tracking.md.

On Activation

Briefly orient the user: this skill implements a Juspay integration from the jp-prd PRD and jp-architecture design. There is no settings file. Then load ./steps/step-01-init.md, which gates on those inputs.

Execution

Read fully and follow: ./steps/step-01-init.md. The step chain:

1. step-01-init.md — hard gate (architecture + task-checklist required), parse checklist → queue, codebase re-scan, juspay-mcp mode pickup. (always)
2. step-02-credentials.md — credentials into .env (provision via juspay-mcp or manual dashboard). (adapts to what's needed)
3. step-03-core.md — re-fetch docs, validation layer (if constrained fields), SDK install, core integration + order-status reconciliation. (always)
4. step-04-webhook.md — webhook handler (signature + idempotency). (only if a webhook task exists)
5. step-05-data-model.md — order/payment schema. (only if a db task exists)
6. step-06-native-setup.md — mobile native SDK setup. (only on native surfaces)
7. step-07-portal-config.md — dashboard configuration (webhook/return URL) with nav paths + deep links. (only if manual-dashboard tasks exist)
8. step-08-test.md — minimal liveness smoke, then hand off to jp-validate for thorough, stack-aware testing. (only if test tasks exist)
9. step-09-summary.md — persistent integration summary; finalize checklist statuses. (always)

Thorough testing lives in the dedicated jp-validate skill (detects the repo's test stack and replicates it, risk-prioritized payment coverage, writes test-report.md). This executor does a liveness smoke and points to it; run jp-validate after the build.

Steps are conditional. Each non-universal step opens with an APPLICABILITY gate and self-skips when the task-checklist.md (authoritative) and architecture.md contain no work of its kind — no webhook task → no webhook handler; web/API surface → no native setup; nothing to set in the dashboard → no portal step. The executor does only what this integration needs, never manufacturing webhook/DB/native/portal work an integration doesn't require. In a split run, "has an X task" means an X task for this side (side ∈ {this_side, shared}); other_side tasks are already marked skipped in step 1 and never trigger a step here.

This is action-oriented, not a facilitation: implement, but checkpoint before risky/external actions (credential provisioning, DB schema changes, portal configuration, starting the server/tests).

juspay-mcp is auth-guarded; not every developer has dashboard access. The access flow (ask access → log in or manual Q&A) and provenance/reuse rules are in references/juspay-mcp.md. Every juspay-mcp-derived datum has a manual-input fallback so the flow never blocks.