* test(3596): adversarial security/prompt-injection abuse suite
Adds `tests/security-prompt-injection.test.cjs` and a fixtures
directory at `tests/fixtures/adversarial/security/` covering the
attack classes enumerated in #3596:
- Command substitution / backticks / heredoc payloads in workstream
names — sentinel-file probes prove no shell is spawned, slugifier
neutralises the input.
- Path traversal through `--ws` and slash-bearing workstream names —
rejected with structured `--json-errors` payload, no stack trace,
no filesystem mutation outside the project root.
- Fake `<system>` / `[SYSTEM]` / `<<SYS>>` / `[INST]` boundary tags —
sanitizeForPrompt neutralises every form; structural negative
property locked across all six styles in one place.
- Zero-width / bidi-override codepoints — stripped per the documented
codepoint set; asserted via codePoint inspection, not regex
literals.
- Hostile read of CONTEXT.md / PLAN.md / ROADMAP.md fixtures —
`gsd-read-injection-scanner.js` surfaces the advisory; excluded
paths and non-Read tools stay silent; malformed JSON does not
crash the hook.
- Hostile write of `.planning/` files — `gsd-prompt-guard.js` emits
a `PreToolUse` advisory; non-Write/Edit tools stay silent.
- Fake `ghp_*` / `sk-*` env tokens — never echoed in CLI stdout or
stderr under hostile inputs; covered under
`// allow-test-rule: structural-regression-guard` because the only
way to assert byte-level absence is `.includes(token)` against the
captured streams.
- `validatePath`, `validateShellArg`, `validatePhaseNumber`,
`validateFieldName` — focused negative-input contract pins.
Pinned behavior gaps (documented, NOT fixed in this PR):
- `<instructions>` is intentionally whitelisted by both the scanner
and the sanitiser (GSD's own prompt scaffolding). Two REGRESSION
GUARD tests lock that contract.
- The current `scanForInjection` does NOT flag malicious markdown
links (javascript:/data:/embedded-credentials URLs). PINNED with
negative-proof so any future scope extension fails the assertion
and forces a deliberate update to the acceptance map.
- `prompt-builder.ts` does not yet wrap plan/context markdown in an
"untrusted data" envelope. That seam lives on the TS side and is
covered by `sdk/src/prompt-builder.test.ts`; out of scope for a
CJS test file. Mentioned in the file header.
Verification:
- `node --test tests/security-prompt-injection.test.cjs` → 73 tests
pass.
- `node scripts/lint-no-source-grep.cjs` → 0 violations across
546 test files (one `allow-test-rule: structural-regression-guard`
annotation on this file for the token-absence assertions).
- `node scripts/run-tests.cjs` → 9730 tests pass, 0 fail.
Refs #3596
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* fix(3596): allow adversarial fixtures in scan + harden graphify status parse
* fix(3596): skip adversarial security fixtures in secret scan
---------
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Three parallel diagnostic agents (G: sdk/install path-sep, H: hook scripts,
I: worktree/workspace) characterised the remaining windows-22 failures on
23b52f1a. Patches by class:
CRLF in test parsers / file-content reads
- bug-2136-sh-hook-version: shebang check split('\n') → split(/\r?\n/)
- bug-3542-executor-git-stash-prohibition: strip \r from read content
(git rewrites stashed text with CRLF on win autocrlf=true checkout)
- workspace: BOM-strip + explicit \r strip in parseCommandFile (BOM at
byte 0 defeats /^---/ anchor → fmMatch null → fm.name undefined)
Windows path-separator / 8.3-shortname normalization
- bug-3491-nested-git-worktree: use fs.realpathSync.native to expand
%TEMP% RUNNER~1 → runneradmin; normalize sep before path-equality
- prune-orphaned-worktrees: normalize \\→/ before substring includes
(git emits forward-slash in --porcelain on Windows even when
path.join produced backslashes)
- bug-3017-codex-hook-absolute-node: accept POSIX path OR path with
drive-letter prefix in hookPath equality assertion
- bug-3126-global-skills-base-runtime-path: use path.join for expected
/xdg/<runtime> values (production calls path.join → \xdg\… on win32)
Test under-specified platform / forgot win32 env
- bug-2979-hook-absolute-node: pass {platform:'linux'} to
buildHookCommand + rewriteLegacyManagedNodeHookCommands so the
POSIX-branch tests don't pick up the #3393 GitBash code path
- bug-3288-model-catalog + bug-3571-config-manifest: also set
USERPROFILE alongside HOME so os.homedir() on win32 redirects to
the test fixture instead of the runner's real ~
External-cmd resolution
- bug-2647-outer-tarball-sdk-dist: use npm.cmd + {shell:true} on win32
so execFileSync resolves PATHEXT (literal `npm` is ENOENT)
ESM loader: tests/runtime-bridge-sync-smoke.test.cjs already migrated to
pathToFileURL in 23b52f1a (cluster E).
Explicit per-test/describe skip on win32 (with required string reasons
to satisfy no-unconditional-win32-skip guard)
- feat-3347-graphify-auto-update-hook: 3 describes — harness spawns
bash/kill/sleep + the hook itself is bash
- feat-3595-fs-fault-injection: move \t and \n filenames into the
POSIX-only branch (NTFS forbids 0x00–0x1F in filenames)
- bug-2775/2829/3033/3231/3359: POSIX shim under ~/.local/bin with
chmod 0o755 — not how Windows install works
- bug-3211: single subtest where cp.execSync reassignment isn't
picked up on win32 (POSIX coverage via the mock; live windows
behavior covered by 3211-D which keeps running)
- install-path-detection: parses sh-style export PATH= rc files;
Windows has no rc files (registry Path)
- worktree-safety-policy: single test using POSIX /repo/wt fixture
paths that can't be expressed under win32 path.resolve
Validated: plex2 (ubuntu docker) 11224/0 pass.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
agents/gsd-planner.md was 49,316 chars after the initial PR; the
planner-decomposition <48K test was passing on main at 49,150 chars (just under
the 49152 limit). My addition pushed it over.
Restructure: instead of teaching the planner agent to read .last-build-status.json
directly, fold the auto-build state into graphifyStatus()'s existing `stale: true`
signal. The planner's existing rule ("if stale: true, treat as approximate") fires
correctly for failed and in-flight auto-builds — no new planner-side prompt content
needed. The full state is exposed under `last_build_auto_update` for callers that
want exit_code / duration_ms / commit-sha context.
- get-shit-done/bin/lib/graphify.cjs: graphifyStatus() reads
.planning/graphs/.last-build-status.json; OR-folds status in {failed, running}
into the existing stale signal; exposes last_build_auto_update field
- agents/gsd-planner.md: revert the auto-update awareness paragraph (49,524 → 49,150)
- agents/gsd-phase-researcher.md: revert the parallel paragraph for consistency
- get-shit-done/references/planner-graphify-auto-update.md: rewrite to document
the graphifyStatus seam instead of planner-side prompt instructions
- tests/feat-3347-graphify-auto-update-config.test.cjs: 4 new graphifyStatus
tests pinning the failed/running/ok/missing matrix
- tests/feat-3347-graphify-auto-update-hook.test.cjs: bump per-spawn timeout
5s → 30s and wait-deadline 5s → 15s to absorb cold-start latency under
parallel-test-file load (full suite runs many *.test.cjs concurrently)
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
