scriptedalchemy/tracedecay/routing-tracedecay

Routing TraceDecay

Use this when you know the kind of question and not the tool. Start with the opening call for that question, then follow only the evidence needed to answer it. Stop when the question is answered. Locate a file before opening it.

A name you already have is a lookup, not a tour of the repository. A question one grep in a file you can already name would answer does not need the graph.

For PR review, read the briefing's ranked entrypoints before widening the diff. Hosted evidence is retained by default; refresh_github: true requests the configured matching PR's provider refresh. Retained CI is not a live inventory of required checks. The brief's output budget is approximate and reports cuts.

diff_context is prospective: it seeds analysis with every indexed symbol in the supplied files. commit_context selects changed declarations from Git: its default is the net HEAD-to-worktree result, including untracked files; staged_only: true selects immutable index blobs. Worktree captures are checked per file, not an atomic checkout snapshot. Read comparison and coverage fields.

git_diff returns structured metadata without hunk bodies. Its working_tree scope means index-to-worktree; use shell git diff or git diff --cached for patch text. source_lines reads exact historical commit/path spans as well as indexed selectors. Continue historical reads at next_line, keeping the original revision, file, and end line. This uses normal admitted source-service availability and path policy. Raw staged blob IDs are not commit selectors.

Report: the answer supported by the returned evidence, plus its freshness line when it is not fresh and any coverage needed to qualify the conclusion. Do not chain calls the evidence did not call for.