Read current requirements content

Read the requirements file content

Read todo file content

Read the verification key. Returns None if it doesn't exist.

Register Anthropic providers from configuration.

Register Databricks providers from configuration. This is async because OAuth authentication requires async operations.

Register embedded providers from configuration.

Register Gemini providers from configuration.

Register OpenAI-compatible providers (e.g., OpenRouter, Groq) from configuration.

Register OpenAI providers from configuration.

Registers all configured providers based on the providers_to_register list. This is an async function because Databricks OAuth registration requires

Reload README.md and AGENTS.md and replace the first system message Returns Ok(true) if README was found and reloaded, Ok(false) if no README was pres

Resolve a commit reference (hash, branch, tag, etc.) to a full hash

Resolve the log file path, trying new path first then falling back to old.

Resolve file paths within a shell command, handling Unicode space normalization. This function finds quoted file paths in a shell command and resolve

Resolve a template variable to its value.

Generate the Soufflé rule body for a single predicate. Returns the clause after `predicate_pass(id) :- `.

Run accumulative autonomous mode - accumulates requirements from user input and runs autonomous mode after each input.

Run a g3 agent in a worktree

Run agent mode - loads a specialized agent prompt and executes a single task. Uses `CommonFlags` for flags that apply across all modes, ensuring cons

Run autonomous mode with coach-player feedback loop (console output).

Run the coach/player implementation loop This function runs the actual implementation phase using g3-core's Agent in a coach/player feedback loop sim

Main entry point for planning mode This function orchestrates the entire planning workflow: 1. Initialize the planning directory 2. Check git status

Run the scout agent and return the research report. This is the blocking part that runs in a background task.

Save the context window to a session file. If session_id is provided, saves to `.g3/sessions/<session_id>/session.json`. Otherwise, saves to `.g3/ses

Save error context to a file for later analysis

Recursively scan a directory for files with given extensions, up to max_depth.

Send keys/text to the element

Set agent mode information for session tracking Called when running with --agent flag to enable agent-specific session resume

Set the session ID (useful for testing)

Set the working directory (useful for testing)

Helper: write a complete plan, approve it, and set up envelope. Returns the actual session ID (which has a unique suffix).

Set up the workspace directory for autonomous mode. Uses G3_WORKSPACE environment variable or defaults to ~/tmp/workspace.

Check if a file should be excluded from staging based on patterns

Split content into frontmatter and body. Frontmatter must be delimited by `---` lines at the start.

Stage files and make git commit

Stamp an envelope with a verification token and re-write it to disk. This is called only when all rulespec predicates pass. It: 1. Gets or creates th

Strip ANSI escape codes from a string. Handles common ANSI sequences like: - CSI sequences: \x1b[...m (colors, styles) - OSC sequences: \x1b]...\x07

Subscribe to completion notifications. Returns None if notifications are not enabled (manager created with `new()`).

Check if the provider supports cache control

Switch to a different session, saving the current one first. This discards the current in-memory state and loads the new session.

Pick a single Unicode indicator for token magnitude (maps to color bands).

Track code fence state (``` markers)

Truncate a query for display

Extract Anthropic-style tool use block

Try to extract from conversation history in session log

Extract feedback from a session log file. Searches backwards for the last assistant message with substantial text content.

Try to extract feedback from session log file

Try to extract the last assistant message from session log (PRIMARY method) This is the preferred extraction method - looks for the last substantial a

Extract OpenAI-style function call

Try to parse a JSON tool call, tracking partial state across chunks.

Unescape backslash-escaped chars: "~/My\ Files" -> "~/My Files"

Update session with PID

Update the .g3/session symlink to point to the given session directory

Validate skill name format. Must be 1-64 chars, lowercase alphanumeric + hyphens, no leading/trailing/consecutive hyphens.

Check if a YAML value contains another value.

Verify the action envelope against the compiled rulespec using datalog. Returns a `VerifyResult` with pipeline stages and verification counts. Stages

Verify a single piece of evidence.

Used in tests, may be useful for future callers

Write the codebase report to discovery directory

Write a human-readable context window summary to file. Format: message_id, role, token_count, indicator, first_120_chars

Write context window summary to file Format: date&time, token_count, message_id, role, first_100_chars

Write the discovery commands to discovery directory

Write a "USER SKIPPED RECOVERY" entry

This test just verifies the enum structure compiles correctly Actual WebDriver tests would require a running browser

Add a cookie

Add a cookie

Check if all verifications passed (no errors or warnings).

Tests for TODO completion detection and file deletion behavior Helper to check if all TODOs are complete (same logic as in lib.rs)