[kylehgc]

Closes #13.

Adopts rtk-ai/rtk#2965 (cherry-pick, authorship preserved) plus one fork-authored amendment and one lockfile chore.

Commits

• 9d678b1 chore: complete Cargo.lock entry for windows-sys dependency — housekeeping from the fix(core): decode process output using Windows console code page rtk-ai/rtk#2717 adoption (PR Adopt rtk-ai/rtk#2717: core - decode process output using Windows console code page #15): the upstream lock hunk omitted windows-sys from rtk's dependency list; cargo build re-resolved it.
• e95207d fix(hooks): don't rewrite pipe-feeding segments whose filter changes output shape — cherry-pick of upstream bc0f568, author XU Pengfei (@xuing). Generalizes the find/fd pipe-skip (Trouble running xargs as second bash command. rtk-ai/rtk#439) into a PIPE_INCOMPATIBLE list (find, fd, grep, rg, ls, head, tail, cat) + is_pipe_incompatible() helper, with 6 new tests. Byte-for-byte faithful: git diff pr-2965 HEAD -- src/discover/registry.rs was empty at cherry-pick time.
• e1e37fd fix(hooks): make pipe-skip normalization match the rewrite path — amendment: the helper's env-prefix heuristic missed sudo/env/quoted env values and absolute binary paths, so sudo grep … | wc -l and /usr/bin/grep … | wc -l were still rewritten (found by pre-PR review, confirmed empirically). Now reuses the same strip_disabled_prefix (ENV_PREFIX) + basename normalization as rewrite_segment, so skip detection and rewriting can never disagree. +4 tests.

Adoption checklist

• Sync: no-op — upstream/develop is already an ancestor of fork develop
• Staleness re-check: develop @ 337bdda still had only the inline find/fd carve-out in rewrite_compound; grep/ls/head/tail/cat/rg pipe-feeders were still rewritten
• Repro on pre-cherry-pick binary (below)
• Cherry-pick preserving authorship; fixed behavior confirmed with real output (below)
• Quality gate: cargo fmt --all clean, cargo clippy --all-targets zero warnings, cargo test --all 2466 passed / 0 failed
• Amendment as separate fork-authored commit

Repro (real output, this machine)

Before (develop @ 337bdda):

$ rtk rewrite "head -50 Cargo.toml | wc -l"
rtk read Cargo.toml --max-lines 50 | wc -l   (exit 3 — rewritten)

$ head -50 Cargo.toml | wc -l                 # raw pipeline
50
$ rtk read Cargo.toml --max-lines 50 | wc -l  # what the hook actually ran
25

wc -l reports 25 for 50 real lines — rtk read elides half the lines. (The installed hook even rewrote one of this session's own verification pipelines mid-investigation, live-demonstrating the corruption.)

After (this branch):

$ rtk rewrite "head -50 Cargo.toml | wc -l"    → exit 1 (passthrough, no rewrite)
$ rtk rewrite "grep -n use src/main.rs | wc -l" → exit 1 (passthrough)
$ rtk rewrite "ls -t | head -1"                 → exit 1 (passthrough)
$ rtk rewrite "sudo grep -n x f.txt | wc -l"    → exit 1 (passthrough — amendment)
$ rtk rewrite "/usr/bin/grep -n x f.txt | wc -l" → exit 1 (passthrough — amendment)
$ rtk rewrite "grep -rn TODO src"               → rtk grep -rn TODO src (exit 3 — savings preserved)
$ rtk rewrite "head -50 Cargo.toml"             → rtk read Cargo.toml --max-lines 50 (exit 3 — savings preserved)
$ rtk rewrite "sudo grep -rn TODO src"          → sudo rtk grep -rn TODO src (exit 3 — savings preserved)

Review

Two-axis review (standards + spec) run pre-PR.

• Spec: faithful cherry-pick, list is exactly the 8 claimed names, all 6 claimed tests present and passing, &&/; segment isolation and quoted-| safety verified in code. One confirmed generalization gap (env-prefix parity claim not actually met for sudo/env/quoted values) — fixed by the amendment.
• Standards: no hard violations; the one Duplicated Code finding (helper reimplementing env-prefix skipping instead of reusing strip_disabled_prefix) is resolved by the amendment, which deletes the duplicate heuristic. Known ceiling: transparent prefixes (e.g. builtin wrapper prefixes the rewrite path strips recursively) are not mirrored in the skip check — same trade-off as upstream, noted for a future ticket if it bites.

🤖 Generated with Claude Code