tamada / oinkie / 34023020512

97%

main: 97%

Pull #104

github

docs: say accurately why lifting is not an MCP tool

Of the three reasons given for leaving lift out, only one still held.

analyzeHeadless exiting 0 on a failed script stopped being an obstacle
in #83, where lift began reading its output back: the failure is now
reported rather than inferred. "Minutes at a time" was asserted and
never measured -- hello_clang lifts in 6.4 seconds -- so the wording
now ties the cost to the binary rather than naming a figure.

The decision is unchanged. Lifting stays out, and you run oinkie lift
yourself, on a host with Ghidra or in the full image. What changes is
that it reads as settled rather than as three obstacles awaiting an
answer, and that it no longer implies a model has to be kept away from
lifting. A model with a shell can run the CLI like anyone else; the
line being drawn is about which step fits inside a request and a
response, not about what may be run.

The same passage lives in four places, two of which the model reads at
runtime -- the server's instructions and oinkie_info's notes -- so all
four move together, and a JSON-RPC session confirms the new wording is
what actually ships.

The link to the lifter's README still pointed at lifter/, which #91
moved to assets/lifters/ghidra/. That is the second reference to that
directory to survive the move because nothing checks it, the first
being the include_str! in src/ghidra/lifter.rs.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

4 of 4 new or added lines in 2 files covered. (100.0%)

5305 of 5487 relevant lines covered (96.68%)

48.33 hits per line