tama adaptive repair

Drafts or accepts a candidate hook script, validates it against the hook's evidence corpus, and stages a repair proposal without changing the live hook.

#Invocation

tama adaptive repair <hook-id> [--patch-file <path>] [--json]

<hook-id> is required. --patch-file validates an operator-supplied candidate; without it, Tama sends the hook's repair brief to a model through Brama, on the route Brama declares for this workshop, and validates the script it answers. --json prints the report as one JSON document instead of key: value lines.

#Output and state

A candidate runs through the interpreter its own #! line names, the way the live hook runs: once for every payload labelled must-block (it has to block, by a non-zero exit, a BLOCKED: line or a {"decision":"block"} line), once for every payload labelled must-pass (it has to exit 0 without a block verdict), and once on {}, where a signal counts as a crash. The report carries ok, checked (the runs made), failures ({kind, ts, detail}) and staged; a report that ran nothing carries only ok and the failure saying why.

Tama prints the report as key: value lines, one per field with nested values compact, or as one JSON document with --json. A valid candidate is copied into a timestamped proposal directory with proposed.diff, report.json, and a summary; the live and archived hook files remain unchanged. The report's ok controls exit status: true exits 0, false exits 1.

#Refusals and failures

• Missing hook id, an id beginning with --, or --patch-file without a value prints adaptive usage and exits 2.
• Unknown or unmapped hook: unknown hook or no script path: <hook-id>.
• Missing candidate: patch file not found: <path>.
• A draft Brama could not give: the reason Brama answered (no declared route, an unreachable gateway, a refused request) or the model on route <route> drafted an empty script, each followed by ; pass --patch-file to propose a script by hand.
• A candidate without a #! line: failure setup, <path> has no #! line naming its interpreter. No labelled cases: failure insufficient-evidence.
• A failed corpus check reports validation did not pass; artifacts left in <tmp-dir>.

Failures that are not the candidate's verdict print doctor: <reason> and exit 1: an interpreter that cannot be started (cannot start <program>: <cause>), a payload that cannot be written to it, a corpus entry without a payload, telemetry evidence that is not intact (naming the segment or lake commit and the check it failed), and a diff that ends with trouble.

Return to the adaptive command group or the complete CLI tree.