Commands

explain

High-confidence assist from cli.error.v0.1 structured JSON.

a11y-assist explain --json <path>

This is the highest-confidence path. When a tool emits structured error JSON following the cli.error.v0.1 schema, explain can anchor its suggestions to the original error ID.

triage

Best-effort assist from raw CLI text.

some-tool 2>&1 | a11y-assist triage --stdin

Parses unstructured text and attempts to detect error IDs. Confidence is Medium if an (ID: ...) pattern is found, Low otherwise.

last

Assist from the last captured log.

a11y-assist last

Reads ~/.a11y-assist/last.log, which is written by assist-run. This is the most common workflow: wrap a command, then ask for help if it fails.

ingest

Import findings from a11y-evidence-engine.

a11y-assist ingest <findings.json> --verify-provenance

Accepts evidence-engine finding files and produces recovery plans from their structured data. The --verify-provenance flag validates that findings have not been tampered with.

assist-run

Wrapper that captures command output for last.

assist-run <cmd> [args...]

Runs the given command, pipes its output to the terminal as normal, and saves a copy to ~/.a11y-assist/last.log. If the command exits non-zero, a hint is printed reminding you to run a11y-assist last.

Shared flags

All commands accept these flags:

Flag	Description
--profile <name>	Select an accessibility profile (lowvision, cognitive-load, screen-reader, dyslexia, plain-language)
--json-response	Output the ASSIST block as JSON to stdout instead of rendered text
--json-out <path>	Write JSON output to a file while still rendering text to stdout