Back to skill

Security audit

postey

Security checks for vulnerabilities and agentic risk

Overview

This social-posting skill needs review because credential-selection flaws can upload media under the wrong identity, and video downloads have incomplete private-network protections.

Prefer a release that fixes local-config trust, project-local OAuth selection and downloader destination checks before using confidential media or untrusted video links. Verify the active identity, avoid repository-supplied credentials and --local OAuth, and isolate remote-video processing from private networks. Install only the intended skill with a reviewed installer version. Keep the documented approval step before publishing or scheduling. The review addressed material scanner gaps locally; hosted-service enforcement was not verified.

Vulnerability Patterns
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (1)

T09 · Insecure Skill Coding Practices

Error
Location
scripts/postey.js:132
Finding

Repository-controlled scope_path bypasses configuration trust checks

Content
View full analysis

Trigger condition

The user runs a media-upload workflow from an attacker-provided repository containing .postey/config.json. No environment credential, usable global OAuth session, or linked CLI token takes precedence.

Impact

Potentially confidential media intended for the user's Postey account is uploaded under an attacker-controlled identity. The account_id used by the surrounding video workflow does not bind the media upload to the intended account.

Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
Findings (21)

Anti-Refusal Statement

High
Category
Anti-Refusal
Confidence
80% 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 · CHANGELOG.md (reported line 250)May include surrounding context.

md
- **The Hermes setup told a headless host to use OAuth, which it can never finish.** `setup.md`
  Step 2 set `auth oauth` unconditionally for every Hermes install, and Step 3's login table
  offered `hermes mcp login postey` with no caveat. A Hermes gateway runs as a background service
  and cannot reach a browser, and there is no `client_credentials` grant — so it registered an
  OAuth client, never received a token, and parked the server. Worse, `auth: oauth` decides the
  method before `headers` is read, so a valid MCP key sitting in the config was never tried. One

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 168)May include surrounding context.

md
Two surfaces exist — MCP tools/resources and the CLI (`postey.js`) — and they are layers, not

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 488)May include surrounding context.

md
- Routing rules (extended): [routing-guide.md](routing-guide.md)

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 · references/mcp-authentication.md (reported line 32)May include surrounding context.

md
| Track | Client | Credential | Header |
|---|---|---|---|
| A, local | anything that can open a browser | OAuth 2.1 access token | `Authorization: Bearer <jwt>` |
| B, web | a web client that added Postey as a connector | none of your own — the connector authorized in Step 2B | — |
| C, headless | CI, containers, cron, SDK callers | MCP key, `mk_*` | `X-API-Key: <key>` |

Anti-Refusal Statement

High
Category
Anti-Refusal
Confidence
80% 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 · references/thread-and-video-formats.md (reported line 23)May include surrounding context.

md
ing:`: body tweet 1 introduces the product with a sourced, verifiable result (cite where it comes from; never invent a first-person test you did not run); tweet 2 contrasts the incumbent on the same sourced task.
- Builder Profile: `This is [Name], [Title] at [Company]. / [N] years ago, [dramatic surprising action]. / The journey: 👇`: body is chronological chapters: origin, first inflection, the bold bet (with a number), the setback, the breakthrough (with a metric), where they are now, one-sentence lesson without moralizing.
- Video Gallery: authentic disbelief opener, lowercase and raw, no "BREAKING:". One clip per post, one sentence of context max; the clip is the post.
- Resource Roundup: `Everyone pays $[Y] for [Z], but [problem]. / I found [N] FREE [things] that [specific outcome].`: one resource per post: name, URL, what it does, how to use it, result.
- Learning Roadmap: `[N] things I learned [doing X] that nobody teaches you:`: each post is one actionable, copy-pasteable step.

Code capability 'env' detected in scripts/postey.js but not covered by declared permissions.

High
Category
MCP Least Privilege
Confidence
75% confidence
Finding

The skill uses 'env' capability that is not covered by the declared 'allowed-tools' preapproval. Review the host runtime's actual enforcement policy; runtime permissions are unknown unless separately verified, and this declaration alone does not establish a permission ceiling.

Content

No source excerpt is available for this finding.

Code capability 'network' detected in scripts/postey.js but not covered by declared permissions.

High
Category
MCP Least Privilege
Confidence
75% confidence
Finding

The skill uses 'network' capability that is not covered by the declared 'allowed-tools' preapproval. Review the host runtime's actual enforcement policy; runtime permissions are unknown unless separately verified, and this declaration alone does not establish a permission ceiling.

Content

No source excerpt is available for this finding.

Cloud Metadata Access

High
Category
Server-Side Request Forgery
Confidence
94% confidence
Finding

The metadata address occurs in a defensive comment, not an active metadata request, but the associated protection is incomplete: assertPublicHttpUrl checks only the literal hostname before delegating fetching to yt-dlp. An attacker-controlled public-looking hostname can resolve to a private address, or redirect there, allowing requests from the agent's network and potentially exposing internal responses; remote cover extraction also passes URLs to ffmpeg without this check.

Content

Scanner excerpt · scripts/postey.js (reported line 706)May include surrounding context.

js
// A remote video URL is fetched by yt-dlp on the user's machine and behind their
// network. Without a check, `video transcribe --input <redacted-url> and
// http://169.254.169.254/ are reachable, and --print "%(title)s" returns the
// fetched page's title into the output JSON — a read primitive, not a blind hit.
// Child processes echo the input URL into stderr, and for a presigned S3 / Drive
// / CDN link the query string IS the credential. This output is read by an agent

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 · scripts/postey.js (reported line 1428)May include surrounding context.

js
const existing  = readConfigFile(localPath) || {};
    // scope_path binds this config to the directory it was written for.
    writeConfig(localPath, { ...existing, scope_path: _resolvedCwd(), oauth: oauthEntry });
    // This file now holds a refresh token, which outlives the access token.
    // cmdSetup guarded its API key this way and the OAuth path never did.
    await _setupGitignore(true, true);
    console.error(fmt.success(`Tokens saved to ${fmt.dim(localPath)}`));

Unsafe Defaults

Medium
Category
Tool Misuse
Confidence
70% confidence
Finding

Tool defaults are unsafe or overly permissive (e.g. disabled TLS verification, no authentication, world-writable permissions). Unsafe defaults widen the attack surface.

Content

Scanner excerpt · CHANGELOG.md (reported line 61)May include surrounding context.

md
subprocesses run without `POSTEY_API_KEY` / `POSTEY_AUTH_TOKEN` in their environment.

**The config file is written atomically at 0600, in a 0700 directory.** Writing in place left an
existing 0644 file world-readable with the key already in it, and followed a symlink if one was in
the way.

**The OAuth callback page escapes what the server sends it**, and only the callback path disarms the

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
75% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · CHANGELOG.md (reported line 98)May include surrounding context.

md
**Publishing guidance no longer contradicts itself.** The hub's automation section waived
confirmation when the user had said "post now" and omitted scheduling entirely; `routing-guide.md`'s
block labelled "Correct Pattern" ended in an ungated `publish_draft`; and `video-workflow.md` gave a
publish recipe with no approval step. All three now state the same rule: explicit yes in the current
turn, on copy the user has actually seen, and scheduling counts as publishing. `video-workflow.md`
also stopped advertising `--schedule` and `--publish-now`, which the CLI rejects outright.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
75% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · CHANGELOG.md (reported line 106)May include surrounding context.

md
`post.trash.list` had no owner; they now belong to the pack that triages failures, which previously
told agents to recreate a post because restore was invisible to it.

**Arming an auto-DM funnel now needs an explicit yes.** `configure_auto_dm` had design rules but no
approval gate, while `reply_comment` had one. An armed automation messages every person who hits the
trigger with no further review, so it now requires the same confirmation, naming the trigger word,
the payload, the account and the post before anything is armed.

The workflow instructs the agent to create a post from transcript-derived content without warning about submission or publication visibility.

Medium
Category
Output Handling
Confidence
88% confidence
Finding

This is an explicit instruction to use MCP create_post, not merely a listing of an available tool. The file presents caption generation as its purpose but does not disclose where the generated content will be submitted or whether the resulting post is a draft, scheduled, or publicly visible.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
88% confidence
Finding

The installation guidance references an unversioned npx package, allowing installation to execute a mutable release of the skills installer. This is a supply-chain hardening issue rather than an unpinned MCP server: a compromised future installer release could execute with the local user's privileges, and selecting only the postey skill does not pin that executable.

Content

No source excerpt is available for this finding.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
80% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · references/mcp-workflows.md (reported line 122)May include surrounding context.

md
`npx skills add posteyai/skills` installs the hub, every optional pack and this repository's skill
template alongside it, which is why every documented form carries `-s postey`.

Never ask the user to paste a large file's contents, and never reach for `read_file` to get media
bytes when a path is missing. It returns metadata and the `cdn_url` by design.

## Video: Choosing a Path

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
93% confidence
Finding

This passage directly recommends installing prerequisites with an unversioned npx skills command, which can execute a changing npm release on the user's machine. It also omits the -s postey selector that this source says prevents installing every optional pack and the repository's template, unnecessarily broadening installed content.

Content

No source excerpt is available for this finding.

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
70% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · scripts/postey.js (reported line 676)May include surrounding context.

js
const probe = spawnSync("ffprobe", ["-v", "quiet", "-print_format", "json", "-show_streams", "-show_format", argv.file], { env: mediaEnv(), encoding: "utf8" });
  if (probe.status !== 0 || !probe.stdout) {
    error("ffprobe failed — is ffmpeg installed?", { hint: "brew install ffmpeg  or  sudo apt install ffmpeg", stderr: redactUrls(probe.stderr)?.slice(0, 300) });
  }
  let data;
  try { data = JSON.parse(probe.stdout); } catch { error("Could not parse ffprobe output"); }

OAuth login claims to honor project-local storage, but authentication ignores the locally stored session.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The --local login path saves an OAuth session to the project configuration, but getOAuthConfig reads only the global configuration and getAuthHeader never consumes local OAuth tokens. Consequently, a successful project-local login can leave subsequent uploads authenticated as a different global identity, potentially disclosing project media to the wrong account; without another credential, authentication simply fails.

Content

No source excerpt is available for this finding.

Unsafe Defaults

Medium
Category
Tool Misuse
Confidence
70% confidence
Finding

Tool defaults are unsafe or overly permissive (e.g. disabled TLS verification, no authentication, world-writable permissions). Unsafe defaults widen the attack surface.

Content

Scanner excerpt · scripts/postey.js (reported line 1494)May include surrounding context.

js
// Write a fresh file at 0600 and rename over the target. writeFileSync's
  // `mode` applies only on create, so writing in place left an existing 0644
  // file world-readable for the window between the write and any chmod — with
  // the key already in it. Rename is atomic, and O_EXCL means we never follow a
  // symlink a repo or a dotfile manager left in the way.
  const tmp = `${configPath}.${process.pid}.tmp`;

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
70% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · video-workflow.md (reported line 15)May include surrounding context.

Linux

pip install yt-dlp sudo apt install ffmpeg

text

For transcription (`video transcribe`), also install Whisper:

Video command help promises draft creation, but the handlers only return inputs for a separate MCP operation.

Low
Category
Excessive Agency
Confidence
99% confidence
Finding

The video post help at L899 promises to create a multi-platform draft, and the transcription help repeats that promise at L942 and L950. In contrast, L625-L648 and L823-L849 deliberately stop after uploading media and return a next_step instructing the caller to invoke MCP create_post, so following the documented commands does not create the promised draft.

Content

No source excerpt is available for this finding.