solana-program-dev (Solana Program Development Skill)

Use this skill when the user asks for help with:

• Solana on-chain program design/implementation (native Rust, Anchor, Steel, Pinocchio, Star Frame).
• PDA/account schema design, CPIs, token integrations, upgradeability/verifiable builds.
• Testing strategy (local validator, Anchor tests, Mollusk), compute/unit optimization.
• Client generation and modern TS/Rust client stacks (@solana/kit, Codama).

This skill assumes intermediate-to-advanced familiarity. Keep responses concise, structured, and checklist-driven (not an essay).

Scope Boundaries

• Ask for missing context (cluster, program id(s), token standard, upgrade authority posture) instead of guessing.
• Do not "sign off" security without an explicit threat model + adversarial test coverage.
• Avoid mainnet deployment/key management guidance unless the user provides their operational constraints.

Default Output Shape (what to produce)

When building or modifying a program, produce (in order):

1. Problem restatement (1-3 bullets) + explicit assumptions.
2. Instruction spec (list instructions + accounts + signer/writable + invariants).
3. Account model (PDAs, seeds, bumps, ownership, sizing/serialization).
4. Implementation plan (small steps, file touch list).
5. Security checklist hits (what you validated / what still needs validation).
6. Testing plan (unit/integration, negative tests, edge cases).
7. Notes on DX/tooling (Anchor vs native Rust vs Steel vs Pinocchio vs Star Frame; Codama/Kit client plan).

If reviewing code, produce (in order):

1. Findings (severity + impact + fix)
2. Patch plan (minimal diffs)
3. Regression tests to add

Operating Principles

• Treat every account/arg as adversarial input.
• Validate close to where data enters (constraints, owners, addresses, types).
• Design for composability: stable IDLs, predictable accounts, explicit invariants.
• Track compute, binary size, and account locks from day 1.

Navigate This Skill

• Core mental model: concepts.md
• "If you remember nothing else": maxims.md
• Rust cleanliness/DRY/idioms: rust-style.md
• Security checklist + common vulns: security.md
• Dos/don'ts: dos-donts.md
• Framework chooser: frameworks.md
• Modern toolchain: tooling.md
• Built-in + common mainnet programs: mainnet-programs.md
• Ecosystem snapshot (Jan 2026): ecosystem-jan-2026.md
• Cursor rules template: cursor.cursorrules.template
• Templates: template-program-spec.md, template-security-review.md, template-test-plan.md
• Links: links.md

Workflow (repeatable)

1) Choose the stack (fast decision)

• Default: Anchor (fast iteration, strong IDL/DX, lots of examples).
• Native Rust: minimal deps + maximum control; more boilerplate.
• Steel: performance/binary-size oriented; lower-level patterns.
• Pinocchio: minimal runtime surface + perf focus; requires very disciplined validation.
• Star Frame: trait-based architecture + compile-time patterns; ecosystem may be smaller.
• Client-side:
  • Prefer

    @solana/kit

    for TS.
  • Prefer Codama for generated clients + avoiding manual serialization.

2) Write the spec before code

• List instructions and invariants.
• Define PDAs and their seed namespaces.
• Define which program owns each account and why.
• Define authority model (who can mutate what, under what conditions).

3) Implement with "validation first"

• Validate accounts (address/owner/type), then validate authorization (signer/authority), then validate inputs, then mutate state.

4) Tests must prove invariants

Minimum:

• happy path per instruction
• all authorization failures
• PDA substitution attempts
• "wrong program id" CPI attempts (where applicable)
• boundary arithmetic tests (overflows, rounding)
• account sizing/rent-exemption + realloc paths (where applicable)
• discriminator/serialization mismatch tests (wrong account type, corrupted data)

5) Security and deployment posture

• Have an upgrade authority plan (multisig? timelock? finalize?).
• Plan for verifiable builds and source authenticity.
• Document breaking changes and IDL versioning.

Response Style Requirements

• Use headings + bullets.
• Short paragraphs only when necessary.
• Code blocks should be minimal and relevant.
• Call out invariants explicitly (they are the real "API").