Merge pull request #95 from LockInTime/t3code/fill-end-of-options

fix(cli): preserve literal fill arguments

Author: yashranaway

.agents/skills/headless-computer-use/references/commands.md

@@ -27,7 +27,8 @@ headless --session NAME inspect --context actions --task "TASK"
 headless --session NAME inspect --context full --text
 headless --session NAME click REF
 headless --session NAME click --role ROLE --name NAME
-headless --session NAME fill REF TEXT
+headless --session NAME fill REF "TEXT"
+headless --session NAME fill REF -- "--json  stays literal"
 headless --session NAME press KEY
 headless --session NAME scroll up|down|top|bottom --amount PIXELS
 headless --session NAME back
@@ -44,6 +45,10 @@ Use `click --role ... --name ...` for unique accessible controls. Use an `@eN`
 ref from the latest inspection when role/name is ambiguous. Inspect again after
 navigation or a large rerender.
 
+Pass fill text as one quoted shell argument so whitespace is preserved. Put
+`--` before a value that contains a literal global flag such as `--json` or
+`--session`; the sentinel itself is not typed into the page.
+
 Use `wait` with the strongest expected condition available:
 
 1. expected URL plus expected text;

CHANGELOG.md

@@ -51,6 +51,11 @@ Cutting that release is tracked in
 - The website now runs on Next.js 16 with its native flat ESLint configuration;
   the legacy `FlatCompat` and `@eslint/eslintrc` path has been removed.
 
+### Fixed
+
+- `fill` preserves quoted whitespace and accepts literal `--json` or
+  `--session` values after the standard `--` end-of-options sentinel.
+
 ### Known gaps
 
 Tracked as [`backlog`](https://github.com/LockInTime/headless/labels/backlog)

apps/headless/Sources/HeadlessProtocol/CLI.swift

@@ -44,9 +44,17 @@ public struct CLIParser {
 
     public func parse(_ rawArguments: [String]) throws -> CLIInvocation {
         var arguments = rawArguments
+        let literalArguments: [String]
+        if let sentinel = arguments.firstIndex(of: "--") {
+            literalArguments = Array(arguments[arguments.index(after: sentinel)...])
+            arguments.removeSubrange(sentinel...)
+        } else {
+            literalArguments = []
+        }
         let jsonOutput = removeFlag("--json", from: &arguments)
         let session = try removeOption("--session", from: &arguments)
         if let session { try validateIdentifier(session, field: "session") }
+        arguments.append(contentsOf: literalArguments)
         guard let command = arguments.first else { throw CLIParseError.missingCommand }
         arguments.removeFirst()
 
@@ -80,11 +88,9 @@ public struct CLIParser {
         case "click":
             return try parseTargeted(.click, arguments: arguments, session: session, jsonOutput: jsonOutput)
         case "fill":
-            guard arguments.count >= 2 else { throw CLIParseError.missingArgument("TARGET TEXT") }
-            let target = arguments[0]
-            let value = arguments.dropFirst().joined(separator: " ")
+            guard arguments.count == 2 else { throw CLIParseError.missingArgument("TARGET TEXT") }
             return remote(.fill, session: session, parameters: [
-                "target": .string(target), "value": .string(value),
+                "target": .string(arguments[0]), "value": .string(arguments[1]),
             ], jsonOutput: jsonOutput)
         case "press":
             guard arguments.count == 1 else { throw CLIParseError.missingArgument("KEY") }
@@ -612,7 +618,7 @@ Commands:
   inspect [--context summary|outline|text|actions|full] [--task TEXT]
           [--within @rN] [--limit N] [--budget TOKENS] [--depth N] [--text]
   click REF | click --role ROLE [--name NAME]
-  fill REF TEXT | press KEY
+  fill REF TEXT | fill REF -- TEXT_WITH_LITERAL_FLAGS | press KEY
   scroll [up|down|top|bottom] [--amount PX]
   back | reload
   wait [--settled] [--url PATTERN] [--text TEXT] [--timeout MS]
@@ -643,6 +649,7 @@ Commands:
 Global options:
   --session NAME   target a named browser session
   --json           emit one JSON object on stdout
+  --               stop parsing global options; quote multi-word fill values
 """
 
 public let capabilitiesDocument: JSONValue = .object([

apps/headless/Tests/HeadlessProtocolTests/ProtocolTests.swift

@@ -299,6 +299,29 @@ struct ProtocolTests {
         )
     }
 
+    static func cliFillPreservesLiteralValue() throws {
+        let value = "pass  --json\tto API"
+        let invocation = try CLIParser().parse([
+            "--session", "qa", "--json", "fill", "@e1", "--", value,
+        ])
+        try expect(invocation.jsonOutput, "global --json before the sentinel should parse")
+        try expect(invocation.request?.session == "qa", "global --session before the sentinel should parse")
+        try expect(invocation.request?.parameters["target"] == .string("@e1"), "fill target should parse")
+        try expect(invocation.request?.parameters["value"] == .string(value), "fill text should preserve whitespace and literal flags")
+
+        let literalFlag = try CLIParser().parse(["fill", "@e1", "--", "--json"])
+        try expect(!literalFlag.jsonOutput, "--json after the sentinel should not become a global option")
+        try expect(literalFlag.request?.parameters["value"] == .string("--json"), "a literal --json fill value should survive")
+
+        let literalSession = try CLIParser().parse(["fill", "@e1", "--", "--session"])
+        try expect(literalSession.request?.session == nil, "--session after the sentinel should not become a global option")
+        try expect(literalSession.request?.parameters["value"] == .string("--session"), "a literal --session fill value should survive")
+
+        try expectThrows("multi-word fill text must stay one shell argument") {
+            _ = try CLIParser().parse(["fill", "@e1", "two", "words"])
+        }
+    }
+
     static func cliSemanticClick() throws {
         let invocation = try CLIParser().parse(["click", "--role", "button", "--name", "Continue"])
         try expect(
@@ -871,6 +894,7 @@ struct ProtocolTests {
             ("command parameter validation", commandParameterValidation),
             ("strict request fields", rejectsUnexpectedRequestFields),
             ("CLI visit", cliVisit),
+            ("CLI fill literal value", cliFillPreservesLiteralValue),
             ("CLI semantic click", cliSemanticClick),
             ("CLI inspect context and task", cliInspectContextAndTask),
             ("CLI conflicting target", cliRejectsConflictingClickTarget),

docs/roadmap/architecture-decisions.md

@@ -258,6 +258,25 @@ Checksums on everything; keep release CI's script-reuse design (the workflow
 calls the same `build.sh`/`test.sh` a developer runs — preserve that
 property when adding PR CI).
 
+## 16. CLI values preserve shell argument boundaries (decided)
+
+**Decision:** global `--json` and `--session` options are recognized only
+before the first `--` sentinel. The sentinel is removed before command
+parsing. `fill` accepts its text as exactly one shell argument rather than
+joining multiple arguments with inserted spaces.
+
+**Status:** decided 2026-08-10 while resolving backlog §A6.
+
+**Rationale:** typed values are data and must reach the browser byte-for-byte
+as represented by the Swift string. Searching the whole argv for global flags
+could silently remove literal text, while joining tokens normalized tabs and
+repeated spaces. Standard shell quoting plus an end-of-options sentinel makes
+the boundary explicit and testable.
+
+**Consequences:** callers quote multi-word fill text and place `--` before a
+value containing a literal `--json` or `--session`. This changes only CLI
+parsing; the wire protocol and protocol version remain unchanged.
+
 ---
 
 ## Decision log
@@ -271,5 +290,6 @@ property when adding PR CI).
 | 8 | Real CDP input on Linux as capability upgrade | Planned (Phase 4) | 2026-08-04 |
 | 12 | Version unification on git tag | Planned (Phase 3) | 2026-08-04 |
 | 15 | Package-manager distribution set | Decided (owner) | 2026-08-04 |
+| 16 | Preserve CLI value boundaries with `--` and shell quoting | Decided | 2026-08-10 |
 
 New decisions append here with the same format.

docs/roadmap/improvements-backlog.md

@@ -92,9 +92,11 @@ Documented in P1 under "Reference lifetime"; covered in
 stripping (`HP/CLI.swift:46-49`) happens before subcommand parsing, so
 `headless fill @e1 pass --json to API` silently drops `--json` from typed
 text; `fill` also joins args with single spaces destroying whitespace
-(`CLI.swift:83-88`). Add a `--` end-of-options sentinel, only strip globals
-before it, and pass the fill value as one argument. Test: fill value
-containing `--json`, tabs, double spaces.
+(`CLI.swift:83-88`). ~~Add a `--` end-of-options sentinel, only strip globals
+before it, and pass the fill value as one argument.~~ **Done:** global options
+are stripped only before the first `--`; `fill` now requires one quoted text
+argument and preserves its whitespace exactly. Protocol coverage includes
+literal `--json`/`--session`, tabs, double spaces, and the quoting boundary.
 
 **A7. Client never verifies response `id`.** ([#18](https://github.com/LockInTime/headless/issues/18)) Failure paths return
 `id:"unknown"` (`HP/Transport.swift:201,222`); `LocalSocketClient.send`