Get a specific research task

Check if the compiled rulespec is empty.

Process a streaming chunk and return completed tool calls if any.

Filters JSON tool calls from streaming LLM content. Processes content character-by-character and removes JSON tool calls while preserving regular tex

Resets the global JSON filtering state. Call this between independent filtering sessions to ensure clean state. This is particularly important in tes

Process an incoming chunk of text. Returns formatted output that's ready to display.

Print a message to stdout

Signal end of stream and flush any remaining content.

Run a shell command and return its output

Compile a rulespec into a datalog-ready representation. This validates the rulespec and converts it into a form that can be efficiently executed by d

Helper to create a chunk

Helper to create a completion chunk

Add a fact to the envelope.

Clear the element's content (for input fields)

Extract facts from an action envelope using the compiled rulespec's selectors. Returns a set of facts that can be injected into datafrog relations.

Classify an error as recoverable or non-recoverable

Create a ToolCall with the given tool name and arguments

Register a new research task Returns the research ID for tracking

Evaluate a predicate against a claim's selected value.

Execute compiled datalog rules against extracted facts. This uses datafrog to evaluate the predicates. The execution model: 1. Create relations for c

Calculate retry delay with exponential backoff and jitter

Helper to create an agent with a mock provider

Helper to create a finished chunk

Create a test agent in a temporary directory

Check if files with given extension exist

Set a default response when queue is empty

Enable native tool calling

Get the current session ID for this agent

Initialize session ID manually (primarily for testing). This allows tests to verify session ID generation without calling execute_task, which would re

Get the path to the g3 binary. In test mode, this will be in the target/debug directory.

Get the g3-plan directory path

Process template variables in the given content. Replaces `{{var}}` patterns with their values. Warns about unknown variables and leaves them unchang

Get the elapsed time since the research started

Execute the `code_search` tool.

Get a theme by name or from file

Create a ToolCall with the given tool name and arguments

Build the interactive prompt string. Format: - Multiline mode: `"... > "` - Plan mode: `" >> "` - No project: `"agent_name> "` (defaults to "g3") - W

Format detailed context statistics as a string.

Check if the result is a success

Apply unified diff to an input string with optional [start, end) bounds. # Arguments `file_content` - The original file content `diff` - The unified

Helper to create a completion chunk

Create a simple text-only response (single chunk + finish)

Generate the stub message content for this fragment.

Acquire the WebDriver session, returning an error message if unavailable.

Parse a provider reference into (provider_type, config_name). Format: "provider_type.config_name" (e.g., "anthropic.default") Falls back to (provider_

Pre-process input to add markdown markers before formatting. ALL CAPS → **bold**, quoted text → special markers for cyan.

Set or clear project content in the system message. Project content is appended to the second system message (README/AGENTS content) so it survives co

Helper to create a completion chunk

Discover all available skills from configured paths. Skills are loaded in priority order (lowest to highest): 1. Embedded skills (compiled into binar

Manually trigger context compaction regardless of context window size Returns Ok(true) if compaction was successful, Ok(false) if it failed

Mint a verification token using a keyed SipHash MAC. The token is computed as: SipHash-2-4(key[0..16], canonical_facts || "\x00" || canonical_rulespe

Save a session continuation artifact This saves latest.json in the session's directory and updates the symlink

Helper to set up a test environment with a temporary directory Returns the temp dir (must be kept alive) and the original directory

Restore the original directory

Calculate the summary max_tokens with provider-specific caps applied. This is the canonical implementation - both manual and auto compaction use this

Get the session directory for a specific session ID. Returns .g3/sessions/<session_id>/

Select values from a YAML value. Returns a vector because wildcards can match multiple values.

Thin context (first third only).

Clip a line to fit within the given width, adding ellipsis if truncated. Uses UTF-8 safe character counting to avoid panics on multi-byte characters.

Update a research task with its result

Format a compiled rulespec and extracted facts as a datalog program. Produces a textual `.dl` file with: - Relation declarations (`.decl`) - Fact ass

Get the repository root (where .git is)

Get the path to the .g3/session symlink

Get the usable terminal width (total width minus right margin). Returns the terminal width minus a 4-character right margin for clarity. Falls back t

--- Public Accessors ---

Check if the rulespec is empty (no claims or predicates).