Back to skill

Security audit

Fiches Clients

Security checks for vulnerabilities and agentic risk

Overview

The skill has a coherent French accounting purpose, but it can automatically make lasting changes to client records and folders, including merges and archives, without enough confirmation or data-minimization safeguards.

Review this skill before installing in a real client workspace. It should require explicit confirmation before merge, archive, reject-draft deletion, slug rename, and bulk document moves, and audit logs should redact or avoid raw sensitive values such as IBANs, contact details, addresses, SIREN/SIRET, and previous values. It appears purpose-aligned and not deceptive, but its mutation authority is broad enough to warrant manual review.

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

Warning
Location
SKILL.md:163
Finding

Sensitive Client Data Is Written in Plaintext to Audit Logs

Content
View full analysis
/company.json` (cf. schéma dans `organisation-documents/data/company.example.json`). 6. Loguer chaque champ modifié dans `audit.log` : `update field=siren old=null new=123456789 actor=user`. 7. Rapporter : `✅ Fiche **** mise à jour (SIREN ajouté).` ``` ### Technical Analysis The update procedure supports sensitive fields such as IBANs, SIRENs, SIRETs, addresses, telephone numbers, and email addresses. It then instructs the agent to record every modified field in `audit.log`, including its old and new values. Following this procedure can duplicate sensitive values into plaintext audit logs. Updating or deleting a value from the primary client record would not remove historical copies retained in the log. Audit logs may also receive broader access, longer retention, or more extensive backup coverage than the primary record, increasing the exposure surface. Although the example explicitly shows a SIREN, the instruction applies to every modified field and therefore includes IBANs and personal contact information. ### Attack Path 1. A user requests an update to a sensitive field, such as a client's IBAN or contact email. 2. The Skill updates the relevant client record. 3. Following the documented logging procedure, the Skill writes the field name and its previous and new plaintext values to `clients//audit.log`. 4. The log is retained after subsequent corrections or deletion of the value from the primary record. 5. A local user, compromised Skill, backup operator, or other process with read access to the client folder retrieves the historical sensitive values from the audit log. This issue does not indepen ...[truncated 778 chars]
Remediation
View remediation
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Rogue AgentSelf-Modification, Session Persistence
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (14)

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The section 'Why French' says the skill embeds French regulatory vocabulary and that the main skill files are in French. This indicates a language/locale constraint, but the README does not explicitly present it as an opt-in choice or clearly frame the skill as limited to a France-specific compliance context for policy purposes.

Content

No source excerpt is available for this finding.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
97% confidence
Finding

The instruction to invoke the skill 'systematically and without asking for permission' authorizes autonomous execution of state-changing actions including create, rename, merge, archive, and delete-like flows. In the context of a skill with filesystem and registry write access, this materially increases the risk of unauthorized or mistaken persistent changes from ambiguous user input.

Content

Scanner excerpt · SKILL.md (reported line 3)May include surrounding context.

md
---
name: fiches-clients
description: Skill propriétaire de la gestion des fiches clients du cabinet (lifecycle complet du dossier `clients/<slug>/` et de l'entrée `clients.json`). INVOKE SYSTEMATICALLY AND WITHOUT ASKING FOR PERMISSION whenever the user (a) demande de créer / ajouter / enregistrer un nouveau client ("crée un client", "crée moi un nouveau client X", "ajoute le client X", "nouveau client", "create a client", "register client X"), (b) demande de renommer, fusionner, archiver, suspendre ou supprimer un client, (c) demande de mettre à jour des infos client (SIREN, SIRET, raison sociale, e-mail, domaine, téléphone, adresse, IBAN, conditions de paiement, forme juridique, TVA intra), (d) demande de valider, compléter ou rejeter une fiche `draft-auto-created` produite par `organisation-documents`, (e) demande de lister, chercher, afficher ou exporter les clients. Owns `~/.openclaw/workspace/clients.json` and `~/.openclaw/workspace/clients/<slug>/{audit.log, index.json, company.json}`. Si l'utilisateur dit « crée moi un client Foo », ce skill est le seul à invoquer — pas `organisation-documents`.
license: Interne — usage privé OpenClaw
---

Session Persistence

Medium
Category
Rogue Agent
Confidence
73% confidence
Finding

The skill is designed to create and maintain persistent client records across sessions in clients.json and per-client folders. Persistence itself is expected for this domain, but when coupled with broad automatic triggers and no confirmation, it can store or alter durable business data based on misinterpreted instructions.

Content

Scanner excerpt · SKILL.md (reported line 3)May include surrounding context.

md
---
name: fiches-clients
description: Skill propriétaire de la gestion des fiches clients du cabinet (lifecycle complet du dossier `clients/<slug>/` et de l'entrée `clients.json`). INVOKE SYSTEMATICALLY AND WITHOUT ASKING FOR PERMISSION whenever the user (a) demande de créer / ajouter / enregistrer un nouveau client ("crée un client", "crée moi un nouveau client X", "ajoute le client X", "nouveau client", "create a client", "register client X"), (b) demande de renommer, fusionner, archiver, suspendre ou supprimer un client, (c) demande de mettre à jour des infos client (SIREN, SIRET, raison sociale, e-mail, domaine, téléphone, adresse, IBAN, conditions de paiement, forme juridique, TVA intra), (d) demande de valider, compléter ou rejeter une fiche `draft-auto-created` produite par `organisation-documents`, (e) demande de lister, chercher, afficher ou exporter les clients. Owns `~/.openclaw/workspace/clients.json` and `~/.openclaw/workspace/clients/<slug>/{audit.log, index.json, company.json}`. Si l'utilisateur dit « crée moi un client Foo », ce skill est le seul à invoquer — pas `organisation-documents`.
license: Interne — usage privé OpenClaw
---

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The invocation trigger is intentionally broad and automatic, covering common phrases for create, rename, merge, archive, update, list, and export operations. In a natural-language environment, overly aggressive matching can cause the skill to run on ambiguous requests and perform persistent mutations without sufficient confirmation.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The skill describes destructive and persistent operations over filesystem records and client registries, but the user-facing behavior minimizes disclosure by emphasizing terse accountant-style messages. That creates a safety gap where users may not realize that a request will rename directories, archive records, or alter shared indexes.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill claims ownership over a narrow set of files, yet its rename/merge procedures also modify shared artifacts such as index-global.json and related path references. That expands its authority beyond the declared boundary, creating a confused-deputy risk where one skill can silently alter global indexing state that other skills rely on.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

Although presented as a client-record manager, the merge flow performs physical movement of all client documents. This crosses from metadata management into document custody, increasing the blast radius of any mistaken invocation and enabling unintended modification of records outside the skill's stated scope.

Content

No source excerpt is available for this finding.

Whitespace Padding

Medium
Category
Prompt Injection
Confidence
70% confidence
Finding

Large whitespace padding was detected (a block of blank lines or a long run of spaces). This can push injected instructions below or to the right of the visible area so a human reviewer never sees them while the agent still reads them. Manual review of the hidden content is recommended.

Content

Scanner excerpt · references/schema-fiche-client.md (reported line 90)May include surrounding context.

md
### `statut`

| Valeur                 | Sens                                                                                            | Transitions autorisées                            |
| ---------------------- | ----------------------------------------------------------------------------------------------- | ------------------------------------------------- |
| `draft-auto-created`   | Créé par `organisation-documents` en auto-création. Attend validation.                          | → `active` (validate-draft), → ∅ (reject-draft)   |
| `draft-user-created`   | Créé par `clients` sur demande user mais marqué incomplet (signaux insuffisants par ex.).        | → `active` (update + validation), → `archived`    |

Whitespace Padding

Medium
Category
Prompt Injection
Confidence
70% confidence
Finding

Large whitespace padding was detected (a block of blank lines or a long run of spaces). This can push injected instructions below or to the right of the visible area so a human reviewer never sees them while the agent still reads them. Manual review of the hidden content is recommended.

Content

Scanner excerpt · references/schema-fiche-client.md (reported line 141)May include surrounding context.

md
## Validation des champs

| Champ          | Règle de validation                                                                                  |
| -------------- | ---------------------------------------------------------------------------------------------------- |
| `slug`         | `^[a-z0-9]+(-[a-z0-9]+)*$`, longueur 1–64                                                            |
| `siren`        | 9 chiffres, validation Luhn FR (algo SIREN)                                                          |

Whitespace Padding

Medium
Category
Prompt Injection
Confidence
70% confidence
Finding

Large whitespace padding was detected (a block of blank lines or a long run of spaces). This can push injected instructions below or to the right of the visible area so a human reviewer never sees them while the agent still reads them. Manual review of the hidden content is recommended.

Content

Scanner excerpt · references/schema-fiche-client.md (reported line 147)May include surrounding context.

md
| `siren`        | 9 chiffres, validation Luhn FR (algo SIREN)                                                          |
| `siret`        | 14 chiffres, commence par `siren`, validation Luhn FR                                                |
| `tvaIntra`     | Format `^[A-Z]{2}[0-9A-Z]{2,12}$`, clé mod 97 pour `FR`                                              |
| `ibanFavori`   | Validation mod 97                                                                                    |
| `codePostal`   | 5 chiffres pour `pays = FR`, libre sinon                                                             |
| `email` (contacts) | Format RFC 5322 simplifié, domaine non générique interdit dans `domains[]` mais autorisé pour un contact |

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
84% confidence
Finding

The description states that the skill owns the full lifecycle of client records, including create, update, rename, merge, and archive operations on clients.json and per-client folders. For a markdown file, this is a data-affecting behavior, but the README does not include a clear warning that using the skill will modify user workspace data and may archive records.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
85% confidence
Finding

The communication rule mandates a specific response style in French ('Style comptable') and gives fixed French output examples, which can amount to forcing a locale/language preference. The file does not indicate that users may choose another language or that the restriction is optional.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Low
Category
Not specified by scanner
Confidence
97% confidence
Finding

La règle générale de garde-fou affirme sans exception que chaque mutation est tracée dans l'audit log du dossier client. La procédure reject-draft contredit explicitement cette règle en imposant un journal global du workspace à la place, ce qui crée une divergence d'intention documentaire sur la localisation de l'audit trail.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
91% confidence
Finding

This JSON example is entirely structured and documented in French, including field comments and sample values such as French legal forms and address conventions. For a general-purpose skill asset, that can constitute a language/locale policy constraint unless the skill is explicitly documented as France-specific or offers user choice.

Content

No source excerpt is available for this finding.