Pine Script v6 — Claude Code Skill

A Claude Code skill for writing, debugging, reviewing, and migrating TradingView Pine Script v6 code.

Its job is to kill the #1 failure mode of LLM-generated Pine: hallucinated APIs (wrong function names, wrong argument order, v5/v6 confusion). It does that by giving the model an offline, queryable reference of every Pine v6 built-in plus a tight set of v6 gotchas — then defers to the live compiler for the final word.

What's inside

pine-script-v6/
├── SKILL.md                      # the skill entry point (workflow + rules)
├── scripts/
│   ├── pine-lookup.mjs           # query the reference without loading 1.8 MB into context
│   └── fetch-pine-docs.mjs       # download pineDocs.json on demand (integrity-checked)
└── references/
    ├── pineDocs.json             # ⤓ NOT committed — fetched by fetch-pine-docs.mjs (see Attribution)
    ├── v6-gotchas.md             # 6 deterministic analyzer rules + common pitfalls
    ├── v5-to-v6-migration.md     # high-frequency v5 → v6 changes
    └── UPSTREAM-LICENSE.txt       # MIT license of the data source

The bundled reference covers 1,114 items: 470 functions, 196 constants, 192 methods, 146 variables, 56 types, 22 operators, 19 control keywords, 10 annotations, 3 fields.

Why a lookup script instead of grep?

references/pineDocs.json is pretty-printed across ~43k lines. Grepping it is noisy and burns context. pine-lookup.mjs parses it once and returns just the entry you asked for — exact signature, typed arguments, return type, see-also, and optionally the official code example.

Installation

Requires Node.js 18+ and Claude Code.

git clone https://github.com/tboome33/pine-script-v6-skill ~/.claude/skills/pine-script-v6
node ~/.claude/skills/pine-script-v6/scripts/fetch-pine-docs.mjs

The second step downloads the built-in reference (pineDocs.json, ~1.8 MB) from the upstream source — it is not committed to this repo (see Attribution). The download is pinned to a specific upstream commit and verified by byte size + SHA-256, so the result is deterministic and tamper-evident. Re-run with --force to refresh.

After that, Claude Code auto-discovers skills under ~/.claude/skills/. The skill triggers automatically when you work on .pine files or mention Pine Script / TradingView indicators, and can also be invoked explicitly with /pine-script-v6.

Prefer a project-local install? Clone into <your-project>/.claude/skills/ instead.

Usage

The lookup script is the workhorse — use it directly or let the skill drive it:

# exact detail for one symbol
node ~/.claude/skills/pine-script-v6/scripts/pine-lookup.mjs ta.ema

# list every member of a namespace (note the trailing dot)
node ~/.claude/skills/pine-script-v6/scripts/pine-lookup.mjs "request."

# substring search across all names
node ~/.claude/skills/pine-script-v6/scripts/pine-lookup.mjs security

# include the official code example
node ~/.claude/skills/pine-script-v6/scripts/pine-lookup.mjs ta.crossover --examples

Flags: --examples, --kind=functions|methods|variables|constants|types|operators|controls|annotations|fields, --json (machine-readable), --names (names only). Exit code 1 when nothing matched, so callers can branch.

Example output:

━━ ta.ema  [Built-in Function]
SYNTAX:
  ta.ema(source, length) → series float
DESC:
  The ema function returns the exponentially weighted moving average...
ARGS:
  • source <series int|float> (required) — Series of values to process.
  • length <simple int> (required) — Number of bars (length).
RETURNS: float
SEE ALSO: ta.sma, ta.rma, ta.wma, ta.vwma, ta.swma, ta.alma

How it fits a compile-verify loop

This skill makes the first pass correct (real signatures in, v6 rules respected). It does not replace a real compiler. If you pair it with a TradingView automation layer (e.g. an MCP server that exposes pine_smart_compile / pine_check), the ideal loop is:

1. Look up unfamiliar built-ins → write code grounded in real signatures.
2. Compile for real → read errors.
3. Fix → repeat until clean.

The skill is the upstream knowledge; the compiler is the downstream source of truth.

The 6 deterministic rules (from v6-gotchas.md)

Ported verbatim from the upstream extension's static analyzer:

1. Array index out of bounds (literal index ≥ size) — error
2. Negative index out of bounds (|index| > size) — error
3. Off-by-one with array.size() used as an index — error
4. Unguarded .first() / .last() on a possibly-empty array — warning
5. Loop bound not derived from the indexed array's size() — warning
6. Implicit int/float → bool cast (v6 forbids it) — warning

Attribution

pineDocs.json is not redistributed in this repository. It is fetched on demand by scripts/fetch-pine-docs.mjs, directly from tradesdontlie/pine-script-v6-extension (MIT — its license text is kept at references/UPSTREAM-LICENSE.txt), pinned to a specific commit and integrity-checked (size + SHA-256). The 6 analyzer rules in v6-gotchas.md are distilled from that project's static analyzer (src/PineStaticAnalyzer.ts). The underlying built-in documentation originates from TradingView's official Pine Script v6 reference, redistributed by the upstream project for offline/educational use. Pine Script® and TradingView® are trademarks of TradingView, Inc. This skill is unofficial and not affiliated with or endorsed by TradingView.

License

MIT — see LICENSE. The fetched pineDocs.json retains its original MIT terms (references/UPSTREAM-LICENSE.txt).