Audit Report: questions — 🟠 D (49/100)

Audited by TAR Engine · 2026-08-09 · Report format v0.2

Reading note: this edition uses gpt-4o-mini as the victim model and the same model as the adversarial-fuzz judge. Findings reflect missing defenses in the SKILL.md itself — not a verdict on any specific victim model. The remediation belongs in SKILL.md, not in the model.

Source: https://github.com/dfrysinger/qrspi-plus/blob/main/skills/questions/SKILL.md

Verdict: High risk — 4 high-severity issues need author attention before deploying to a shared environment.

What this skill does

Auditor's read (LLM-generated): The "questions" skill generates targeted research questions based on an approved goals.md, ensuring that the questions do not reveal any goals or intents. It categorizes questions into types such as [codebase], [web], and [hybrid] to guide the appropriate research efforts, while isolating the research process from the goals. The skill also includes mechanisms for review and approval of the generated questions before they are finalized.

Author description: Use when goals.md is approved and the QRSPI pipeline needs research questions generated — produces tagged questions that guide the Research step without leaking goals

Observed: questions is 8 top-level sections (Overview, Artifact Gating, Execution Model, Process, Red Flags — STOP, …); ~179 lines of instructions, delegates to packaged scripts, concise body.

Frontmatter facts:

• Body size: 179 lines / 14319 chars

Score breakdown by category

Each category gets its own sub-score. A category with no rule hits gets 100; a category with a single critical finding drops to 80.

Historical baseline (same-skill comparison)

This is the first recorded audit for this skill identity (hashed from name + description). The baseline section will show mean / stddev / trend after 2+ audits accumulate.

Findings

7 rules matched. Each finding below cites the matched line and a remediation hint.

1. 🟠 SEM-008 — external_payload_blind_trust (HIGH)

• Category: Malicious payload signatures
• Why this matched: This process assumes that the diff content is safe and does not validate it, which could lead to executing malicious code if the content is manipulated.
• Rule intent: Trusts external content (downloaded file, remote prompt template, third-party output) without validation
• Matches in document: 1

Evidence (1 of 1 match):

Line 83:

The orchestrator runs `git -C "<repo>" diff "<ref>" -- "<ABS_ARTIFACT_DIR>/questions.md" > "<ABS_ARTIFACT_DIR>/reviews/questions/round-NN.diff" as a Bash redirect.

Suggested fix: Implement validation checks on the diff content before processing it to ensure that it does not contain harmful or unintended commands.

2. 🟠 SEM-002 — ambiguous_instruction (HIGH)

• Category: Prompt injection / scope override
• Why this matched: This instruction could be interpreted differently based on the context of the user input, potentially allowing a user to bypass the restriction if they manipulate the input.
• Rule intent: Ambiguous instruction that could be exploited as a prompt injection primitive
• Matches in document: 1

Evidence (1 of 1 match):

Line 28:

Do NOT generate questions without an approved goals.md.

Suggested fix: Clarify the instruction by specifying the exact conditions under which questions can be generated, and consider implementing a check to enforce this rule strictly.

3. 🟠 SEM-004 — prompt_injection_subtle (HIGH)

• Category: Prompt injection / scope override
• Why this matched: The phrasing of the question could be subtly manipulated by an adversarial user to extract sensitive information or influence the model's output.
• Rule intent: Subtle nudge to hide, mislead, or impersonate that regex couldn't catch
• Matches in document: 1

Evidence (1 of 1 match):

Line 50:

BAD: "We want to add real-time notifications — how do competitors handle this?" (leaks the goal)

Suggested fix: Rephrase the examples to avoid any language that could be interpreted as revealing intent or goals, and ensure that all generated questions are strictly neutral.

4. 🟠 SEM-007 — irreversible_action_no_confirmation (HIGH)

• Category: Shell safety
• Why this matched: Automatically approving changes without explicit user confirmation can lead to unintended consequences if the changes are not thoroughly vetted.
• Rule intent: Skill instructs the LLM to take an irreversible action without explicit user confirmation
• Matches in document: 1

Evidence (1 of 1 match):

Line 109:

When `config.md` carries `pipeline: quick`, the human-approval gate is skipped after any review round.

Suggested fix: Require explicit user confirmation before skipping the human-approval gate, regardless of the pipeline configuration, to ensure that all changes are reviewed.

5. 🟡 SEM-006 — credential_handling_unsafe (WARNING)

• Category: Credential exposure
• Why this matched: If user feedback includes sensitive information, writing it to a file without proper handling could expose credentials or personal data.
• Rule intent: Credentials/tokens handled in ways that risk exposure (logging, echoing, including in prompts)
• Matches in document: 1

Evidence (1 of 1 match):

Line 105:

write the user's feedback to `feedback/questions-round-{NN}.md`

Suggested fix: Ensure that any sensitive information is sanitized before being written to files, and consider implementing a secure handling mechanism for user feedback.

6. 🟡 SEM-003 — capability_overreach (WARNING)

• Category: Prompt injection / scope override
• Why this matched: Accessing configuration files may grant the skill unnecessary permissions that could be exploited if the configuration is not properly secured.
• Rule intent: Capability claim over-broad relative to the skill's stated purpose
• Matches in document: 1

Evidence (1 of 1 match):

Line 25:

Read `config.md` from the artifact directory to determine whether Codex reviews are enabled.

Suggested fix: Limit the skill's access to only the necessary configuration settings and ensure that sensitive configurations are protected from unauthorized access.

7. 🔵 QL-001 — shell_block_no_error_handling (INFO)

• Category: quality
• Why this matched: Shell block missing set -e / || exit — silent failures will go unreported
• Rule intent: Shell code blocks without set -e or explicit error handling
• Matches in document: 1

Evidence (1 of 1 match):

Line 87:

     86: 
>>   87: ```sh
>>   88: REVIEW_STEP="questions"
>>   89: REVIEW_ROUND="${ROUND}"                                  # current review round (NN)
>>   90: REVIEW_OUTPUT_DIR="<ABS_ARTIFACT_DIR>/reviews/questions/round-${ROUND}/"
>>   91: REVIEW_ARTIFACT="questions.md"
>>   92: REVIEW_AGENTS="quality-claude=qrspi-questions-reviewer,quality-codex=qrspi-questions-reviewer"
>>   93: ```
     94: 

Suggested fix: Add set -euo pipefail at the top of bash blocks, or chain critical commands with || exit 1. Skills that fail silently mid-script are nearly impossible to debug downstream.

Scope of this edition

The audit covers static rule matching, semantic-layer LLM analysis, and adversarial prompt fuzzing. Three classes of risk live beyond this edition's scope. We name them explicitly:

• Runtime behavior. Verifying what a skill actually does at runtime requires sandboxed execution. That layer ships in a future edition; today's report reflects what the skill states it will do, plus the LLM's read of how it would behave.
• Cross-skill composition. When this skill is chained with others through a planner, the emergent state flow between skills is its own analysis surface. Out of scope for single-skill reports.
• External payloads. A skill that fetches and runs a remote script is flagged at the fetch step. The remote payload itself is audited as a follow-up once the sandbox layer is online.

Methodology

How the score was computed:

1. Document text is scanned against a static rule set of 32 signature patterns. Each rule carries a permanent rule_id (e.g. PI-001), a category, a severity, and a remediation template.
2. Each rule hit deducts from a 100-point base: critical -20, high -10, warning -5, info -1.
3. The letter grade is gated by max severity AND total score: any critical → F; any high → at most D; any warning → at most C; otherwise A/B by score band.
4. Per-category sub-scores apply the same deduction formula to that category's findings only — so you can see WHICH risk surface drove the loss.

Rule matches are augmented by an LLM-based semantic pass when an LLM endpoint is configured. The semantic pass uses rule IDs SEM-001 … SEM-008.

When an LLM endpoint is configured the skill is also probed with a 15-attack adversarial corpus (5 classes × 3 prompts), each judged by a separate LLM call. Failed classes surface as rule IDs AR-001 … AR-005.

Engine + rule set provenance:

• Engine version: 0.2.0
• Rule set version: 1.1.0
• Commit: unknown
• Domain config: general
• Audited at: 2026-08-09T20:29:09.365507Z
• Rules applied: 36 static rules (full registry below)

Known limitations of this report

• False positives are possible. A SKILL.md documenting a dangerous pattern (e.g. an audit skill explaining curl | sh) will match the rule even though the skill's intent is to detect, not execute. Read the matched lines before reacting.
• False negatives are guaranteed in narrow ways. Patterns obfuscated by string concatenation, environment variable indirection, or non-English equivalents will slip past regex.
• Baseline sample size. Same-skill trend analysis (§ Historical baseline) gets meaningful with n≥3 prior audits. With fewer priors the stddev band is widened to avoid false out-of-band signals.

About TAR Engine

TAR Engine is an OSS "wish machine" with built-in audit. Speak a goal; the engine plans, runs and audits skills inside its own container. BYOK. — github.com/qingxuantang/tar-engine