Back to plugin

Security audit

Solaris Dev Shop

Security checks for vulnerabilities and agentic risk

Overview

Review is needed because some bundled database workflows can affect the wrong database or delete more history than described.

Keep database workflows away from production until the WordPress import target and revision-retention commands are corrected. Stop database clients before SQLite restoration, use private temporary files, and explicitly approve external services and persistent changes. Installation itself does not activate these operations. Scanner coverage was partial; this targeted review establishes concrete concerns without certifying every bundled workflow.

Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Memory PoisoningPersistent Context Injection, Context Window Stuffing, Memory Manipulation
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • Rogue AgentSelf-Modification, Session Persistence
Findings (1117)

YARA rule 'php_webshell_known': Known PHP webshell families (c99, r57, b374k, WSO, etc.) [webshells]

Critical
Category
YARA Match
Confidence
90% confidence
Finding

YARA rule matched a known webshell pattern (PHP, Python, JSP, or ASPX webshell).

Content

Scanner excerpt · employees/specialized/wordpress-master/diagnostics/diagnostic-malware-scan/SKILL.md (reported line 131)May include surrounding context.

and reachable behavior. Perform only separately authorized quarantine or restoration after confirmation.", "confidence": "heuristic" }

text

## Category 2: Known Backdoor Signatures (Critical)

These patterns match known backdoor and webshell signatures.

**Pattern 2a: Common backdoor names (case-insensitive)**
```bash
find -P "$SCAN_ROOT" -type f -name '*.php' -exec grep -Hni -- 'FilesMan\|WSO\|c99shell\|r57shell\|b374k' {} +

What it finds: References to well-known PHP backdoors and webshells.

Pattern 2b: Shell execution with user input

bash
find -P "$SCAN_ROOT" -type f -name '*.php' -exec grep -Hn -- 'shell_exec\s*(\s*\$_\|system\s*(\s*\$_\|passthru\s*(\s*\$_' {} +

What it finds: Direct execution of shell commands from user input ($_GET, $_POST, $_REQUEST). Major security risk.

For each match in Category 2:

Finding ID: SUSP-BACKDOOR-{3-char-md5-of-filepath:line}

Severity: Critical

Category: Suspicious Code

Example Finding:

json
{

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · HARNESS-BUILD.json (reported line 3647)May include surrounding context.

json
"executable": false,
      "sha256": "822ea7427cfccb3867a488972f2b0b4187977508aca0f81b889b3c22f335a0c2"
    },
    "examples/capabilities/G22/fixtures/test-secret.txt": {
      "executable": false,
      "sha256": "075af0498293cf0e94ef9127b173318c74260f2260950d35e56f41231a2de361"
    },

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · employees/specialized/payments-specialist/webhook-ledger.md (reported line 6)May include surrounding context.

md
`payment_ledger.py` is a working local model of the part of a store backend that takes payment
webhooks: orders, captures, partial refunds and cancels, with exact money and conserved stock. It
never contacts a payment provider. Events are local JSON files, signed with a local test secret
(`examples/capabilities/G22/fixtures/test-secret.txt` is a demo value that grants nothing). Python
3.9+ stdlib, one SQLite file. It is the reference behaviour to port into a real handler, not a
replacement for the provider's own idempotency or for reconciliation against the provider's records.

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · employees/specialized/payments-specialist/webhook-ledger.md (reported line 41)May include surrounding context.

bash
python3 employees/specialized/payments-specialist/payment_ledger.py init --db ledger.db --catalog examples/capabilities/G22/fixtures/catalog.json
python3 employees/specialized/payments-specialist/payment_ledger.py sign --secret-file secret.txt --body-file event.json --t 1790000000
python3 employees/specialized/payments-specialist/payment_ledger.py ingest --db ledger.db --secret-file secret.txt --events stream.jsonl --now 1790000200
python3 employees/specialized/payments-specialist/payment_ledger.py report --db ledger.db --json
python3 employees/specialized/payments-specialist/payment_ledger.py check --db ledger.db

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · employees/specialized/payments-specialist/webhook-ledger.md (reported line 42)May include surrounding context.

bash
python3 employees/specialized/payments-specialist/payment_ledger.py init --db ledger.db --catalog examples/capabilities/G22/fixtures/catalog.json
python3 employees/specialized/payments-specialist/payment_ledger.py sign --secret-file secret.txt --body-file event.json --t 1790000000
python3 employees/specialized/payments-specialist/payment_ledger.py ingest --db ledger.db --secret-file secret.txt --events stream.jsonl --now 1790000200
python3 employees/specialized/payments-specialist/payment_ledger.py report --db ledger.db --json
python3 employees/specialized/payments-specialist/payment_ledger.py check --db ledger.db
python3 employees/specialized/payments-specialist/payment_ledger.py export-g14 --db ledger.db --out books/ --period-start 2026-09-01T00:00:00Z --period-end 2026-10-01T00:00:00Z

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · examples/capabilities/G22/README.md (reported line 3)May include surrounding context.

md
# G22 payment refunds and inventory conservation: fixtures, oracle, controls, demo

- `fixtures/catalog.json`: a synthetic store (3 SKUs, USD). `fixtures/test-secret.txt`: a local demo
  signing value that grants nothing. `fixtures/september.jsonl`: 13 signed envelopes written by
  `make_stream.py` (fixed timestamps; ingest with `--now 1790000200`).
- `ledger_oracle.py [--tool PATH] [--seeds N]`: 10 checks against its own reference model and the G14

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · examples/capabilities/G22/README.md (reported line 30)May include surrounding context.

md
S=employees/specialized/payments-specialist/payment_ledger.py
F=examples/capabilities/G22/fixtures
python3 $S init --db $T/ledger.db --catalog $F/catalog.json
python3 $S ingest --db $T/ledger.db --secret-file $F/test-secret.txt --events $F/september.jsonl --now 1790000200 > $T/first.json || echo "ingest exit $? (3: two envelopes refused)"
python3 $S ingest --db $T/ledger.db --secret-file $F/test-secret.txt --events $F/september.jsonl --now 1790000200 > $T/replay.json || echo "replay exit $?"
python3 $S report --db $T/ledger.db
python3 $S check --db $T/ledger.db

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · examples/capabilities/G22/README.md (reported line 31)May include surrounding context.

md
F=examples/capabilities/G22/fixtures
python3 $S init --db $T/ledger.db --catalog $F/catalog.json
python3 $S ingest --db $T/ledger.db --secret-file $F/test-secret.txt --events $F/september.jsonl --now 1790000200 > $T/first.json || echo "ingest exit $? (3: two envelopes refused)"
python3 $S ingest --db $T/ledger.db --secret-file $F/test-secret.txt --events $F/september.jsonl --now 1790000200 > $T/replay.json || echo "replay exit $?"
python3 $S report --db $T/ledger.db
python3 $S check --db $T/ledger.db
python3 $S export-g14 --db $T/ledger.db --out $T/books --period-start 2026-09-01T00:00:00Z --period-end 2026-10-01T00:00:00Z || echo "export exit $? (1: evt_007 and evt_008 are open)"

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · examples/capabilities/G22/ledger_oracle.py (reported line 50)May include surrounding context.

python
self.tool, self.tmp = tool, tmp
        self.db = tmp / (name + ".db")
        (tmp / (name + "-cat.json")).write_text(json.dumps(catalog))
        (tmp / "secret.txt").write_text(SECRET.decode() + "\n")
        r = self.run("init", "--db", self.db, "--catalog", tmp / (name + "-cat.json"))
        if r.returncode != 0:
            raise AssertionError("init failed: %s" % r.stderr[:200])

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · examples/capabilities/G22/ledger_oracle.py (reported line 62)May include surrounding context.

python
def ingest(self, lines, now=T0 + 1000, *extra):
        f = self.tmp / ("%s-batch-%s.jsonl" % (self.db.stem, uuid.uuid4().hex))
        f.write_text("\n".join(lines) + "\n")
        r = self.run("ingest", "--db", self.db, "--secret-file", self.tmp / "secret.txt", "--events", f, "--now", now, "--tolerance", "100000", *extra)
        out = json.loads(r.stdout) if r.stdout.strip().startswith("{") else None
        return r.returncode, out

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · tests/test_g22_capability.py (reported line 22)May include surrounding context.

python
EX = ROOT / "examples" / "capabilities" / "G22"
FIX = EX / "fixtures"
ENV = dict(os.environ, PYTHONDONTWRITEBYTECODE="1")
SECRET = (FIX / "test-secret.txt").read_text().strip().encode()
NOW = 1790000200

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · tests/test_g22_capability.py (reported line 56)May include surrounding context.

python
p.write_text("\n".join(path_or_lines) + "\n")
        else:
            p = path_or_lines
        r = run(TOOL, "ingest", "--db", self.db, "--secret-file", FIX / "test-secret.txt", "--events", p, "--now", now, *extra)
        return r.returncode, (json.loads(r.stdout) if r.stdout.strip().startswith("{") else None)

    def report(self):

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · SOLARIS-BUILD-SOURCE.json (reported line 920)May include surrounding context.

json
"examples/capabilities/G22/README.md": "8dbb01aa1ce30a1c760409fb2d3ce843470a27cc668b939d405ee27734c44034",
    "examples/capabilities/G22/fixtures/catalog.json": "741b07a581d4b5d3798f14dec88aa099e4081df526c0d3734d58e9d19506faeb",
    "examples/capabilities/G22/fixtures/september.jsonl": "822ea7427cfccb3867a488972f2b0b4187977508aca0f81b889b3c22f335a0c2",
    "examples/capabilities/G22/fixtures/test-secret.txt": "075af0498293cf0e94ef9127b173318c74260f2260950d35e56f41231a2de361",
    "examples/capabilities/G22/ledger_oracle.py": "eff8064d78adcd8ab012b4d59b598968fa09728c15edb38da6cebf7113b6196b",
    "examples/capabilities/G22/make_stream.py": "18d82f05473465ba2eb797e9f1063c4d36ce87d4bfe9e2d188cbf3b70bef32e3",
    "examples/capabilities/G22/naive_ledger.py": "40881706b7884474e1f42fe77b146d7f391e9687d29aad6a31e34c141b0f74f4",

Credential Access

High
Category
Privilege Escalation
Confidence
60% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · chief-of-staff/capability.contract.json (reported line 122)May include surrounding context.

json
"uncontrolled self-modification of contracts, budgets, or roster",
   "auto-absorb external skills without security tiers and human review",
   "perform specialist craft when role is coordinator",
   "place any credential, API key, client secret, or .env value into an assignment packet or handoff brief; packets carry the secret's location reference only",
   "emit a packet that authorizes an irreversible act (production deploy, DNS/registrar change, data or repo deletion, client-visible send, payment) without the owner's confirmation quoted in the packet",
   "cancel or rewrite an already-emitted assignment record; assignments are append-only and only the owner cancels"
  ],

Credential Access

High
Category
Privilege Escalation
Confidence
60% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · employees/design/technical-writer/capability.contract.json (reported line 141)May include surrounding context.

json
to a live/public surface (Mintlify, Read the Docs, GitHub Pages, a client docs domain, llms.txt on a public host): requires human confirm; drafts stay on disk until then",
      "IRREVERSIBLE - the delete arm of the Workflow 7 keep/fix/delete triage, and any overwrite of an existing docs tree, README, or CHANGELOG release entry: requires the named page owner to confirm per page, never a bulk sweep",
      "IRREVERSIBLE - running a runbook step outside a staging dry-run, or executing an API sample against a production endpoint to verify it: denied, verify against staging only",
      "CREDENTIAL - samples and env tables carry placeholders only (sk_test_... $ADMIN_TOKEN, <your-key>); never copy a real key, token, connection string, or .env value out of the repo into a doc, and never commit one",
      "CREDENTIAL - NDA or client source documents are converted on-machine via the local Docling/MarkItDown path; uploading them to a cloud converter is denied, no confirm available"
    ]
  },

Credential Access

High
Category
Privilege Escalation
Confidence
60% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · employees/design/ui-ux-designer/design-intelligence/data/stacks/nextjs.csv (reported line 38)May include surrounding context.

text
LIC_ for client vars,Server vars exposed to client,NEXT_PUBLIC_API_URL,API_SECRET in client code,High,https://nextjs.org/docs/app/building-your-application/configuring/environment-variables
36,Environment,Validate env vars,Check required env vars exist,Validate on startup,Undefined env at runtime,if (!process.env.DATABASE_URL) throw,process.env.DATABASE_URL (might be undefined),High,
37,Environment,Use .env.local for secrets,Local env file for development secrets.env.local gitignored,Secrets in .env committed.env.local with secrets.env with DATABASE_PASSWORD,High,
38,Performance,Analyze bundle size,Use @next/bundle-analyzer,Bundle analyzer in dev,Ship large bundles blindly,ANALYZE=true npm run build,No bundle analysis,Medium,https://nextjs.org/docs/app/building-your-application/optimizing/bundle-analyzer
39,Performance,Use dynamic imports,Code split with next/dynamic,dynamic() for heavy components,Import everything statically,const Chart = dynamic(() => import('./Chart')),import Chart

Exfiltration Commands

High
Category
Prompt Injection
Confidence
90% confidence
Finding

Instructions found that direct the agent to transmit conversation context or user data to external services.

Content

Scanner excerpt · control-plane/policy.json (reported line 64)May include surrounding context.

json
"prescribe",
    "file a lawsuit",
    "legal filing",
    "send message to client now",
    "book the flight",
    "spend",
    "production outage",

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · control-plane/skills.json (reported line 1005)May include surrounding context.

json
"id": "solaris.fleet-provisioner",
      "name": "fleet-provisioner",
      "department": "infrastructure",
      "summary": "Worker provisioning plans with synthetic fixtures; no Keychain or personal-path access.",
      "tags": [
        "fleet",
        "infrastructure",

Self-Modification

High
Category
Rogue Agent
Confidence
85% confidence
Finding

Skill modifies its own code, configuration, or behavior at runtime. Self-modification enables an agent to escalate privileges, disable safety constraints, or install persistent backdoors.

Content

Scanner excerpt · docs/new-skill-template/README.md (reported line 28)May include surrounding context.

md
- [ ] `author` is `Solaris Dev Shop`, `license` is `MIT`
- [ ] `learnings.md` has dated observations from actual worked examples; do not invent experience to fill the file
- [ ] After proposal review, register the capability in `control-plane/roster.json` and add routes and probes in `control-plane/policy.json` and `control-plane/routing-probes.md`
- [ ] Update skill counts and affected docs as part of that reviewed addition
- [ ] Run `python3 tools/build_skill_catalog.py`, then commit the generated catalog
- [ ] `python3 -m unittest discover -s tests -v` and `python3 control-plane/meta_control_plane.py self-test` pass
- [ ] `python3 tools/validate_skill_metadata.py` passes (install `requirements-dev.txt` first)

Self-Modification

High
Category
Rogue Agent
Confidence
90% confidence
Finding

Skill modifies its own code, configuration, or behavior at runtime. Self-modification enables an agent to escalate privileges, disable safety constraints, or install persistent backdoors.

Content

Scanner excerpt · employees/README.md (reported line 19)May include surrounding context.

md
- Receives gap-requests from the learnings sweep when "we keep lacking X"
- Reports back with external solutions to absorb per employee

Improvement is reviewed work, not an automatic background loop or self-updater.

## Structure

Anti-Refusal Statement

High
Category
Anti-Refusal
Confidence
85% confidence
Finding

Skill instructs the agent to omit warnings, disclaimers, or ethical commentary. Stripping safety caveats hides risk from the user and is a common jailbreak preamble.

Content

Scanner excerpt · employees/creative-media/3d-artist/rules.md (reported line 40)May include surrounding context.

md
- Scale not 1 unit = 1 m (asset imports giant or tiny in-engine)

## Debugging moving rigs
- When an imported/retargeted animation looks twisted, mirrored, flat, offset, or foot-sliding, run the structured motion-state inspection in blender-motion-debugging.md - extract facts first (scene inventory, skeleton, forward/up axes, sampled frames, contact diagnosis), screenshots only to confirm. Do not judge a moving rig from one camera angle.

## What this employee does NOT do
- Engine logic / gameplay / scenes / builds (Unity / Unreal developers)

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
94% confidence
Finding

The deletion is ordinary cleanup, but the surrounding workflow creates and consumes a predictable filename in shared /tmp. Where another local process can pre-create or replace that path, the initial redirection can follow a symlink and overwrite a writable file, or FFmpeg can consume a substituted manifest; concurrent legitimate runs can also interfere.

Content

Scanner excerpt · employees/creative-media/video-editor/SKILL.md (reported line 212)May include surrounding context.

bash
printf "file '%s'\n" file1.mp4 file2.mp4 file3.mp4 > /tmp/concat_list.txt
ffmpeg -n -f concat -safe 0 -i /tmp/concat_list.txt -c copy "$OUTPUT"
rm /tmp/concat_list.txt

Different codecs (re-encodes)

Anti-Refusal Statement

High
Category
Anti-Refusal
Confidence
70% confidence
Finding

Skill instructs the agent to never refuse or to always comply. Suppressing the agent's ability to decline removes a core safety control and enables downstream harmful requests to succeed.

Content

Scanner excerpt · employees/data/data-analyst/rules.md (reported line 9)May include surrounding context.

md
ion depends on an event, don't track it (sickn33 analytics-tracking).
- **Metric definitions are contracts.** Every metric carries: exact formula, data source, refresh frequency, owner, target + threshold, and a decision rule "if below X, then Y" (voltagent data-analyst + alirezarezvani product-analytics).
- **Profile data quality first** - before base queries, before charts (voltagent implementation order; alirezarezvani DQS gate).
- **Statistically significant and practically significant are separate questions. Always answer both** (alirezarezvani statistical-analyst).
- **Segments reveal lies blended numbers tell.** Always segment by cohort, plan tier, channel, geography before believing an aggregate (alirezarezvani product-analytics anti-patterns).
- **Tell the story, don't dump numbers.** Headline = [Specific Number] + [Business Impact] + [Actionable Context] (wshobson data-storytelling).
- **Explain findings in business terms, not statistical jargon** (lodetomasi data-detective).

Anti-Refusal Statement

High
Category
Anti-Refusal
Confidence
80% confidence
Finding

Skill attempts to nullify the agent's safety policies or restrictions ('you have no restrictions', 'ignore your guidelines', 'do anything now'). This is a direct jailbreak that disables guardrails.

Content

Scanner excerpt · employees/data/data-analyst/rules.md (reported line 23)May include surrounding context.

md
e × effect size × practical impact; check the risk triggers before any verdict.
- **When** experiment design, power analysis, causal inference, Bayesian/bandit questions arise → hand to Data Scientist. Analyst reads results; Scientist designs tests.
- **When** a dashboard is requested → stakeholder interview first: which decisions, which audience altitude, which refresh cadence. Decisions-backward design.
- **When** exploratory SQL must run against production → read replica or hard LIMIT; never heavy queries on prod without safeguards (sickn33 sql-pro safety).
- **When** the same ad-hoc query is written a third time → flag to Data Engineer for promotion to an owned dbt model.
- **When** data needed doesn't exist → file an instrumentation gap with the event spec (name, properties, trigger, decision supported) - don't approximate silently.
- **When** comparing periods → use comparable windows (WoW, MoM) and same-period-last-year for seasonal businesses (alirezarezvani product-analytics).

Memory Manipulation

High
Category
Memory Poisoning
Confidence
90% confidence
Finding

Skill manipulates agent memory, state, or stored context. Memory corruption can alter personality, override safety rules, or cause unpredictable behavior.

Content

Scanner excerpt · employees/data/data-engineer/ingestion-orchestration-lakehouse-quality.md (reported line 38)May include surrounding context.

md
k (query/restore a prior snapshot), hidden partitioning (partition without leaking partition columns into every query - kills the "forgot the partition filter" full-scan class), in-place schema AND partition evolution without rewriting data, row-level MERGE/upsert on the lake.
- **Catalog decision**: a catalog tracks table metadata pointers - REST catalog (engine-neutral, increasingly the default), AWS Glue, Snowflake-managed, or Polaris. Pick one catalog as the single source of truth; multiple uncoordinated catalogs over the same files corrupt state.
- **Maintenance is real work**: schedule compaction (small-file problem), snapshot expiration, and orphan-file cleanup - an unmaintained Iceberg table degrades read performance and leaks storage. Treat these as pipeline tasks with the same observability bar.
- **Iceberg vs Delta**: both are mature; choose by engine ecosystem (Iceberg = broadest multi-engine + Trino/Flink; Delta = Databricks-first). Conform to the client's existing stack.

Endor dependency reachability

Checked Nov 8, 2026 · 7:26 PM UTC

No reachable vulnerabilities reported

Endor dependency scan completed.