loadPolicy() read .llm-security/policy.json from whatever root it was
given, and every scanner passes the SCANNED TARGET: scan-orchestrator
(policyRoot = resolve(args.target)), entropy-scanner (thresholds and
suppression patterns), signature-scanner (sig.custom_rules_path and
enabled_families), trigger-scanner (phrase lists) and ast-taint-scanner
(enabled, python_path). A foreign/cloned target could raise its own
entropy thresholds, disable SIG families, supply its own SIG ruleset or
name the interpreter the AST scanner spawns — configuring the scan of
itself. Same defect class as S3b's .llm-security-ignore fix.
Chosen: move isOwnWorkingTree() to scanners/lib/own-working-tree.mjs (one
copy, reused by the orchestrator's ignore-file check) and make
loadPolicy() refuse an EXPLICIT root that is not the caller's own tree —
defaults plus one stderr line, same form as S3b — because one rule in one
function covers every scanner and a future call site cannot forget it.
The IMPLICIT root (CLAUDE_PROJECT_ROOT/cwd, what every hook uses) is the
caller's own project by construction and is read as before.
entropy-scanner's calibration.policy_source no longer reports an ignored
file as its source.
New tests/scanners/policy-scope.test.mjs was red on 0d37f5a (foreign
target: entropy finding silenced, custom SIG rule loaded, findings differ
from the same tree without policy.json, no stderr line) and is green now;
its own-tree scenario (known-positive) is green before and after. The 15
existing policy tests that placed own-tree fixtures under os.tmpdir() now
use tests/helpers/own-tree.mjs (fixture under $HOME, cwd set to it).
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The five Python taint fixtures under tests/fixtures/ast-scan/ move into
tests/helpers/payload-trees.mjs as the `ast-scan` tree, written to a temp
dir by the test like the three S1 trees. av-surface probe (d) gains the
directory: 5 -> 6 trees.
PM decision (S3 order): creds-net.py is os.environ -> requests.post, the
exfiltration shape AV classifiers are trained on; one rule, "no
payload-shaped runnable file on disk", is easier to defend than an
exception. The .py files match no SIG rule and are stored as plain lines.
Deviation from the order: it named three files (sinks, creds-net, scope);
the directory holds five (also reassign.py, sentinel.py). Gating the
directory means all five move.
Measured: (d) red first, d=1 (5 files) of 6 trees, then 0. sha256 of all
five on-disk files taken before git rm; payload-trees.test.mjs asserts
them, mutation-checked (one byte in creds-net.py -> red, restored
byte-identical). ast-taint-scanner.test.mjs 12/12 with the materialized
tree, both before and after the files were removed.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
py-ast-taint.py Assign handler never removed a name from the tainted set on reassignment, so a source-then-constant/sanitizer rebind (g=os.getenv(); g='safe'; os.system(g), or x=shlex.quote(x)) kept stale taint and fired false AST-CMD-EXEC findings. Assign now clears taint for target names when the RHS is not a source. No cross-expression propagation added — the f-string/concat/alias recall gap (#27) stays deferred to v8.
#28: added fixtures+assertions locking the tainted subprocess/os.system AST-CMD-EXEC sink and the open(...,'w') AST-FILE-WRITE sink. Suite 1931/0.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TcQyMTQfyrsAapaCMPxTtQ
