Format a remember tool result summary.

Format a take_screenshot result summary.

Format token count in compact form (e.g., 1K, 2M, 100b, 200K) Clamps to 4 chars right-aligned.

Format a workspace path for display, replacing home directory with ~.

Generate a unique fragment ID.

Generate a histogram showing tokens used and wall clock time per turn.

Get agent-specific language prompt for a specific agent and language.

Get an embedded agent prompt by name.

Get the path to the envelope.yaml file for a session.

Get the path to the .g3 directory

Get the most recent fragment ID for a session (the tail of the linked list).

Get stop sequences for a model type.

Get the path to the todo.g3.md file. Checks for G3_TODO_PATH environment variable first (used by planning mode), then falls back to todo.g3.md in the

Get tool definitions including any dynamically loaded toolsets.

Check if all items are terminal (done or blocked).

True if position is at start of text or preceded only by whitespace after newline.

Check if a process is running by PID

List all pending (not yet injected) research tasks

Load a fragment from disk.

Load configuration with CLI argument overrides applied. This is the canonical function for loading config with CLI overrides. All CLI entry points sh

Log an error message to the session JSON file as the last message This is used in autonomous mode to record context length exceeded errors

Mark session as paused (for interactive sessions)

Merge new notes into existing memory. Appends new notes to the appropriate sections or creates new sections.

Parse a selector string into a Selector. Examples: - `csv_importer.capabilities` -> [Field("csv_importer"), Field("capabilities")] - `tests[0].name`

Parse commit message from LLM response

Parse a streaming chunk from Gemini's SSE response Gemini streams JSON objects, one per line (not SSE format with "data:" prefix)

Parse a unified diff into a list of hunks as (old_block, new_block). Each hunk contains the exact text to search for and the replacement text includin

Print a g3 status message with colored tag and status Format: "g3: <message> ... [status]" Uses centralized G3Status formatting.

Print a compact Plan tool output with styled content Format: " ● plan_read" or " ● plan_write" header, then styled plan items plan_yaml: None for no p

Remove a worktree and its branch

Remove the header line (# Workspace Memory and > Updated: ...) from content.

Reset the JSON tool call filter state. Called at the start of a new response to clear any partial state. Default implementation does nothing.

Try to resolve a file path, handling Unicode space normalization. This function attempts to find a file in the following order: 1. Try the path as-is

Restore session from a continuation artifact Returns true if full context was restored, false if only summary was used

Retry logic for async operations

Run interactive mode with console output. If `agent_name` is Some, we're in agent+chat mode: skip session resume/verbose welcome, and use the agent na

Save session metadata

Save the entire context window to a per-session file

Create a temporary directory with a test file

Check if thinking mode should be disabled for a summary request. Anthropic requires: max_tokens > thinking.budget_tokens + 1024

Feed content in random-ish chunk sizes

Get the element's text content

Truncate text at the first stop sequence, if any.

Truncate a string for display, adding ellipsis if needed.

Extract JSON tool call pattern

Try to extract a complete JSON object from the buffer Gemini streams responses as a JSON array: [{...}, {...}, ...] We need to handle the array brack

Validate the claim structure.

Validate a provider reference (format: "<provider_type>.<config_name>")

Validate that the system prompt is the first message in the conversation history. This is a critical invariant that must be maintained for proper agen

Create a copy of the config with a different default provider

Write a plan to the session's plan.g3.md file.

Internal function to accept a session with a custom prefix

Check if all diagnostics passed

Apply cyan highlighting to quoted text markers (runs after markdown formatting).

Build the user message for requirements refinement This message instructs the LLM to read the codebase and refine requirements.

Build a GeminiRequest from a CompletionRequest.

Calculate max_tokens for a summary request.

Check if plan is terminal and exit plan mode if so. Returns true if plan mode was exited (plan is complete or all blocked).

Check startup state and determine if recovery is needed

Check if buffer matches the tool call pattern. Pattern: `{` followed by optional whitespace, `"tool"`, optional whitespace, `:`, optional whitespace,

Force a token refresh by clearing any cached token This is useful when we get a 403 Invalid Token error

Clear the session continuation symlink (for /clear command) This only removes the symlink, not the actual session data

Clear the conversation history while preserving system messages. Used by /clear command to start fresh.

Collect all pending output into a single string.

Generate the completed requirements filename

Generate the completed todo filename

Compress a shell command to fit within the given width. Preserves the command name and as much of the arguments as possible. Truncates from the right

Convert g3 tools to Gemini format

Convert Gemini usage metadata to g3 Usage

Count how many cache_control annotations exist in the conversation history

Try to decode bytes as UTF-8, handling incomplete sequences at the end. Returns the decoded string and leaves any incomplete bytes in the buffer.

Determines which providers should be registered based on mode and configuration. In autonomous mode, registers coach and player providers in addition

Discover skills from configured paths and generate the skills prompt. Returns the skills prompt section if any skills are found, None otherwise. Skil

Internal helper: thin all context and track the event

Emit a complete blockquote.

Emit formatted inline content.

Emit a complete table.

Ensure both llvm-tools-preview and cargo-llvm-cov are installed Returns Ok(true) if tools were already installed, Ok(false) if they were installed by

Escape a string value for use in a datalog literal. Replaces backslashes, double quotes, and newlines with escape sequences.

Estimate token count for messages.

Estimate token count from text (~4 chars per token)