Get the file path for this fragment.

Find a child element by CSS selector

Find multiple child elements by CSS selector

Find a fallback title from the first few lines of content.

Find a matching open delimiter in the stack.

Find the first incomplete stage (for resumption)

Flush any incomplete constructs.

Flushes any pending content from the JSON filter. Call this at the end of streaming to ensure any buffered newlines or whitespace that wasn't followe

Manually trigger context thinning regardless of thresholds

Manually trigger context thinning for the ENTIRE context window Unlike force_thin which only processes the first third, this processes all messages

Format a code_search result summary.

Format a compact summary for a successful compact tool

Format seconds as human-readable duration

Format milliseconds as a human-readable duration string.

Format an action envelope as pretty YAML with comments.

Format a header line.

Format only inline code within text (used for nested formatting in links)

Format a load_toolset result summary.

Format messages according to the model's native chat template.

Format nested bold within italic text

Format nested italic within bold text

Format a plan_approve result summary.

Format a plan as markdown with embedded YAML.

Format a session's created_at timestamp for display

Format the file size in a human-readable way.

Format a byte size for display.

Format a timestamp for filenames Format: YYYY-MM-DD_HH-MM-SS (filesystem-safe)

Format tool output lines for display, respecting machine vs human mode.

Determine how to format a tool result for display.

Format verification results as a string for display. Uses loud formatting for warnings and errors. If session_id is provided, also checks that envelop

Format a YAML value as indented markdown.

Generate a git commit message based on the requirements Uses the planner LLM to generate a commit summary and description. Returns (summary, descript

Generate a summary of requirements for planner_history.txt Uses the planner LLM to generate a concise summary of the requirements. The summary is at

Generate a short unique ID (first 8 chars of UUID v4).

Generate a summary of the pipeline run

Get Anthropic config by name

Get a map of all available agents (both embedded and from workspace).

Get the path to the background processes directory. Returns `.g3/background_processes/` in the workspace or current directory.

Get the provider reference for coach mode in autonomous execution

Get the current git branch name

Get Databricks config by name

Get an embedded skill by name.

Get the path to the errors directory. Returns `.g3/errors/` in the workspace or current directory.

Get the modified time of a file as a formatted string

Get the current HEAD commit hash

Generates initial discovery messages for fast codebase exploration. This function: 1. Runs explore_codebase to get a codebase report 2. Sends the rep

Get the path to the latest.json continuation file This follows the symlink to get the actual path

Get the path to the memory file. Memory is stored at `analysis/memory.md` in the working directory (version controlled).

Get OpenAI config by name

Read or create the verification key at `~/.g3/verification.key`. - If the key file exists, reads and returns it. - If it doesn't exist, generates 32

Get or initialize the global llama.cpp backend

Get the provider reference for player mode in autonomous execution

Get the default LLM provider

Get all requests that were made to this provider

Get stop sequences based on model type.

System prompt for providers without native tool calling, with skills support.

Get only the text content before the first JSON tool call. Returns the full buffer if no tool calls are found. This is used to save the "preamble" tex

Get the tools output directory for a session. Returns .g3/sessions/<session_id>/tools/

Returns chunks to send for a content_block_delta event.

Returns chunks to send for a content_block_start event.

Returns chunks to send when a content block finishes.

Handle a character in Buffering state

Handle a control command. Returns true if the command was handled and the loop should continue.

Handle a detected delimiter.

Handle execution errors with detailed logging and user-friendly output.

Handle a newline character.

Handle recovery situation

Handle a character in Streaming state

Handle a character in Suppressing state (string-aware brace counting)

Check if any content was loaded.

Check if conversation history has consecutive assistant messages. Returns the indices of the first consecutive pair found, if any.