pilfer

Decrypt all ansible vault files in a project in-place recursively for viewing/editing, then re-encrypt them all at once when you're done.

Optionally decrypt/re-encrypt all encrypted variables in-place, or re-key an entire tree after a password exposure.

Output

Example session:

$ pilfer open --include-encrypted-vars --quiet
🔓 Searching for and decrypting vault files...
ℹ️  Found 4 vault target(s) (3 whole-file, 1 with inline encrypt_string)
✅ All vault files decrypted. Edit as needed, then run 'pilfer close' to re-encrypt.
$ pilfer close
🔒 Re-encrypting vault files...
ℹ️  Re-encrypting modified file: inventory/group_vars/all/secrets.yml
ℹ️  Re-encrypting modified inline vault string(s) in: inventory/group_vars/all/all.yml (1 changed)
✅ Vault files re-encrypted. 2 modified files have been updated.

Quick start

Requires Python 3.10+ and Ansible on PATH.

pipx install pilfer
cd your-ansible-project
pilfer open
# edit or search plaintext
pilfer close

Unchanged files are restored to their original ciphertext automatically.

Add to ansible.cfg so you do not need -p on every run:

[defaults]
vault_password_file = ~/.ansible-vault/.vault-file

Features

• ansible.cfg integration - Automatically reads vault_password_file from your ansible.cfg
• Change detection - Only re-encrypts files that were actually modified (using SHA256)
• Safe operation - Preserves original encrypted content for unchanged files
• Ansible vault - Calls Ansible's VaultLib. pip install pilfer installs Ansible; a checkout needs Ansible on PATH
• Binary data preservation - Preserves exact line endings and formatting (critical for certificates)
• Inline encrypt_string support - Opt-in via pilfer open --include-encrypted-vars; decrypts YAML !vault scalars in place (with # pilfer:vault:N markers); close always re-encrypts whatever the session opened
• Fail-closed sessions - Refuses double-open, keeps session state if close partially fails, non-zero exit codes on errors

Usage

pilfer [--version] COMMAND ...

Commands: open | close | rekey | set

pilfer open [--include-encrypted-vars] [-q] [-p VAULT_PASSWORD_FILE]
pilfer close [--confirm-delete] [-p VAULT_PASSWORD_FILE]
pilfer rekey --old-vault-password-file OLD --new-vault-password-file NEW [options]
pilfer set [-p VAULT_PASSWORD_FILE] FILE KEY [VALUE]

Run pilfer --help or pilfer COMMAND --help for full options and examples.

Set a single value

Use pilfer set to add or replace a top-level key in a whole-file vault without opening a session:

pilfer set secrets.yml db_password

Pilfer prompts for the value without echoing it. You can also pipe a value in:

printf '%s' "$TOKEN" | pilfer set secrets.yml db_password

Piped input drops one trailing newline, so echo works too. You can pass the value as the final argument, but it may appear in shell history and process listings. Put -- before a value that starts with -.

The value stays in memory and is saved as a string, even if it looks like a YAML boolean or number. New keys are appended; replacing an existing scalar preserves any inline comment on its line. Values containing {{, {% or {# are tagged !unsafe so Ansible uses them as typed. An existing !unsafe tag is kept when the new text would still load as a string. !unsafe "0123", !unsafe "yes", and !unsafe dates are stored as plain quoted strings, because Ansible would otherwise retype them.

Pilfer refuses plaintext files, files in an open Pilfer session, and key names that YAML reads as a boolean or null, such as yes, on and null. If the operation fails, the original ciphertext remains unchanged. Pilfer replaces the file while preserving its mode; hard links continue to point to the old content.

Set values from Python

Scripts that already hold a secret can use the in-process API to keep it out of command arguments:

from pilfer.vault_set import set_keys

set_keys(path, {key: value}, password_file)

Failures and refusals raise PilferError, including an unreadable password file or failed encryption.

To check whether an update can be made before an action you cannot undo, such as minting a token, call set_keys with a placeholder value and check_only=True. This performs the edit and encryption and writes a temporary ciphertext file beside the vault, but skips the final rename. It can therefore catch problems such as a read-only directory or full disk.

Inline encrypted variables (encrypt_string / !vault)

Whole-file vaults are opened by default. Inline !vault scalars are opt-in:

# Open whole-file vaults AND inline encrypt_string values
pilfer open --include-encrypted-vars

# Edit values in place. pilfer rewrites each !vault block like:
#   db_password: "the-secret"  # pilfer:vault:0
#
# Do NOT remove the `# pilfer:vault:N` comment - close uses it to find
# and re-encrypt each value. Do NOT commit while those markers are present
# (plaintext secrets + session metadata would land in git).

pilfer close   # no flag needed; re-encrypts everything this session opened

close always re-encrypts session entries (whole-file and inline). The --include-encrypted-vars flag is only meaningful on open.

If you delete an entire opened variable line (key + value + marker), close refuses by default (ambiguous delete vs accident). Confirm with:

pilfer close --confirm-delete

which then prints:

🔍 Detected removal of 1 encrypted vars:
  - db_password

If you strip only the # pilfer:vault:N comment while leaving the secret, close refuses with:

⚠️  inventory/group_vars/all/all.yml — cannot close (marker missing)
  Variable:     db_password
  Problem:      # pilfer:vault:0 marker was removed but the secret is still in the file.

  Fix:          restore the marker comment on the value line, then pilfer close

If you delete the whole opened variable line (key + value + marker), close refuses by default (ambiguous delete vs accident):

⚠️  inventory/group_vars/all/all.yml — cannot close (secret line deleted)
  Variable:     db_password  (# pilfer:vault:0)
  Problem:      opened inline secret was removed from the file.

  Confirm delete:  pilfer close --confirm-delete
  Undo:            restore the line, then pilfer close

Renaming the key and dropping the marker is also refused if the secret value is still present in the file (including in comments).

Vault password file

Pilfer finds the vault password in this order: -p, then vault_password_file in ansible.cfg, then ~/.ansible-vault/.vault-file, .vault_password, vault_password_file, and finally ../../vault_password_file.

Examples

pilfer open
pilfer open -p ~/.my-vault-password
pilfer open --include-encrypted-vars
pilfer close

Re-key a project

To preview a password rotation across the current directory and its subdirectories, run:

pilfer rekey --old-vault-password-file OLD --new-vault-password-file NEW --dry-run

This includes inline !vault values by default. See Rotating the vault password for the full procedure.

Rotating the vault password

pilfer close is not password rotation - it refuses a different password than the one used for open (anti re-key). To rotate every vault target in the tree (including inline !vault spans):

# Plan / decrypt-check only
pilfer rekey \
  --old-vault-password-file ~/.ansible-vault/.vault-file \
  --new-vault-password-file /tmp/new-vault-pass \
  --dry-run

# Re-key ciphertext (prompts: type REKEY). Inline spans included by default.
pilfer rekey \
  --old-vault-password-file ~/.ansible-vault/.vault-file \
  --new-vault-password-file /tmp/new-vault-pass

# After 100% success, optionally archive the old password file and install the new
# one at the old path (chmod 600):
pilfer rekey ... --rotate-password-file

Refuse to rekey while a pilfer session is open. Nested git checkouts are skipped (run pilfer rekey from those directories separately). Prefer --dry-run first. A mid-run failure can leave a split-password tree; re-run the same rekey command to resume (files already on the new password are skipped). --rotate-password-file is refused with --no-include-encrypted-vars, and also when nested git checkouts were skipped, so the live password file is not rotated while ciphertext remains on the old password. Stale .pilfer-rekey-* staging files are ignored as vault targets and removed only after confirmed mutating rekey (never on --dry-run).

Safety

Pilfer fails closed: if it cannot prove a secret is safely re-encrypted or intentionally removed, it keeps the session and .vault/ backups and exits non-zero. It does not invent fixes for ambiguous edits.

Failure modes this protects against

• Silent re-key on close with a different password than open
• Double-open destroying encrypted backups under .vault/
• Orphan plaintext after deleting vaultedFileList.json (markers, .vault/, or *.pilfer-open sidecars still block re-open)
• Stranded plaintext after stripping markers, renaming keys, or relocating secrets (including into comments)
• Crash mid-decrypt leaving unmarked plaintext (open sidecars are written before plaintext)

Surprising-by-design behaviors

• Interrupted close retries: whole-file targets only count as already done when working bytes match the open backup, or the file is vault ciphertext decryptable with the session password (not an arbitrary foreign vault blob).
• close is progressive - files that succeed are encrypted and dropped from the session; failures stay plaintext until you fix and retry. Not an all-or-nothing transaction.
• Intentional var removal requires pilfer close --confirm-delete.
• Short secrets can block close if the same bytes appear elsewhere in the file (docs/comments) - fail closed.
• Nested git checkouts are skipped; run pilfer from those roots if needed.
• Legacy unbound sessions can close only if the password decrypts the session backups (then pilfer binds a v2 fingerprint); otherwise remove the session list and re-open.
• Incomplete open (session list written, crash before decrypt) is cleared on the next open only when listed paths still look like vault ciphertext and there are no backups/sidecars - so you are not told to close ciphertext. If the session list remains but files are already plaintext (artifacts deleted), open still refuses.
• *.pilfer-open sidecars sit beside opened files (whole-file opens have no # pilfer:vault: markers).

Gitignore

vaultedFileList.json
.vault/
**/*.pilfer-open
*.pilfer-lock

Pre-commit hook (suggested)

Block commits while a session is open:

# .git/hooks/pre-commit (chmod +x)
if [ -e vaultedFileList.json ] || [ -d .vault ] \
  || find . -name '*.pilfer-open' -print -quit 2>/dev/null | grep -q .; then
  echo "pilfer session open (vaultedFileList.json / .vault / *.pilfer-open); run pilfer close first"
  exit 1
fi
# Optional: also refuse # pilfer:vault: markers from --include-encrypted-vars
if git grep -n '# pilfer:vault:' -- '*.yml' '*.yaml' >/dev/null 2>&1; then
  echo "files still contain # pilfer:vault: markers; run pilfer close first"
  exit 1
fi

Recovery

• Session present (vaultedFileList.json) → fix the reported issue → pilfer close again.
• Session deleted but markers / .vault / *.pilfer-open remain → restore vaultedFileList.json from backup if you have it and close, or manually re-encrypt / restore secrets before open.

License

GPLv3+. See PILFER_LICENSE.txt.

Credits

Borrows heavily from the excellent, but no longer supported Ansible Toolkit.

Maintainers: development setup, releases, and CI details in .github/workflows/README.md.