Extends the Phase A-tier-1 site-side rotation engine with an optional Node
browser-sidecar and a deterministic self-heal tier, all proven LIVE-VM against
the local lab change-password form. go-rod stays the default; the sidecar is a
swap-in Driver behind the same Session/Driver/Rotator interfaces — no spine
change, backup + verify-new-before-revoke-old + proof gates all apply unchanged.
M-B1 sidecar parity: internal/browserrot/sidecar/ (index.js JSON-RPC over stdio,
patchright real-fingerprint Chromium, playwright-core fallback) + sidecar.go
Driver/Session. A shared test harness runs go-rod and the sidecar through the
identical live rotation proof.
M-B2 stealth: patchright removes navigator.webdriver; humanized input (Bezier
mouse paths + per-key typing jitter, no ML trajectory generation per plan §5).
M-B3 self-heal: Healer interface + heal-retry fill/click/text wrappers; on a
rotted selector, heal.js relocates the field from DOM structure ONLY (never a
value) and the engine rewrites + persists the recipe (recipes.go RecipeBook)
so the next run is Tier-1 again. Deterministic heuristic is the proven Tier-2;
Stagehand LLM strategy is opt-in + UNPROVEN. Leak-check asserts no secret ever
reaches a heal request (redacted IPC transcript).
Secret seam documented honestly: fill carries base64 secret bytes across exactly
one local stdio boundary, typed and wiped, never logged/persisted/sent to a
model. MFA/CAPTCHA still degrade to the human worklist (Hard Rule 5).
14 tests green, -race clean; lab/lab-provision-browserrot.sh proves both drivers.
M-B4 (first REAL self-owned site) deliberately NOT started — needs per-credential
authorization (safe-candidate ladder).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Rotate one Gitea PAT and rewrite it in EVERY on-disk reference in lockstep,
so a real in-use token can rotate without breaking git auth. The blob carries
one ref=<file> per reference; Rotate mints the new token, proves it
authenticates, then rewrites the old->new token literal in each ref file
(atomic temp+rename, minGiteaTokenLen-guarded), Verify asserts the new token
is present in each ref, and the spine revokes the old token last. A dead token
is never written into a git config; any failure leaves every ref on a live token.
Adds internal/rotate/gitea_refs.go (read-only reference enumerator) + driver and
enumerator unit tests. Proven end-to-end against real Gitea 1.25.0 in the sandbox
VM via lab/lab-gitea-blast-vm.sh (GITEA_BLAST_VM_OK): one rotation rewrote a fake
.git-credentials, an embedded git-remote URL, and a tea config; new->200, old->401.
Two documented follow-ons before any in-use host PAT: gopass-entry-sync (PAT
duplicated across other gopass entries) and token-scope-cloning (preserve the old
token's repo scopes on mint, like sendgrid).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
First real-credential rotation on the safe-candidate ladder: full
create→verify→store→revoke against a real Gitea PAT on
git.churchofmalware.org (self-owned, no 2FA). To keep blast radius at
zero, the run targets a purpose-made THROWAWAY token — a genuine token
used by nothing — so the driver's real lifecycle earns LIVE-REAL without
risking any in-use PAT (which is also referenced across .git-credentials,
embedded git remotes and tea config, and needs the multi-reference
blast-and-cutover, still to build).
- lab/lab-rung3-gitea-real.sh: env-only secrets, isolated throwaway gopass
store, independent cutover check (new->200, old->401), sealed backup.
- proofs.go + ROTATION-PROOFS.md: gitea promoted LIVE-VM -> LIVE-REAL,
LIVE-REAL section added; "no driver carries this yet" removed.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
First end-to-end run of the real rotate --execute spine on the host toolchain —
real gopass, mandatory backup gate, audit log, INCREDIGO_PASSPHRASE seal flow —
against a THROWAWAY scratch app (dummy .env + config.yml, fake SECRET_KEY) in an
isolated store. Walks dry-run -> blast -> execute, asserts the in-place cutover
(old value gone from both files, one identical new value written, stored blob
updated, sealed backup produced) and that the old value is recoverable from the
backup. Confirms the safety-relevant asymmetry: only plan/dry-run/blast scan the
host (read-only, noop); --execute sources ONLY gopass entries under --prefix, so
it can rotate nothing but the scratch entry.
appsecret stays LIVE-VM: the scratch secret is fake, so this proves the host flow,
not a real credential. Promotion to LIVE-REAL is reserved for a real-cred rotation
with explicit per-credential authorization.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The CLI test proves export→import logic against a fake gopass; this proves the
real recovery path end-to-end against real gopass in an isolated throwaway store:
seed → seal → destroy every entry → import → assert byte-equal restoration and no
plaintext in the bundle. It is the safety net for no-op-RevokeOld drivers (e.g.
appsecret), where the sealed .age backup is the only rollback — the prerequisite
for the first real rotation.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Align banner-A/B/C/D with the corrected hero claim: drop the false
"leaves no trace on disk" wording (gopass stores GPG ciphertext on
disk) in favor of "no plaintext at rest — GPG-encrypted on disk".
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Deterministic, headless browser-driven password rotation:
discover change page via RFC 8615 / links, inject old+new over CDP
(no model in the loop), submit, and re-login to verify the new
password before commit. Implements rotate.Rotator so the mandatory
backup gate, verify-before-revoke ordering, and proof gate apply
unchanged; RevokeOld is a no-op (the site invalidates the old pw).
Proof-as-data per Site: the engine is LIVE-VM (real headless
Chromium vs a real local change-password form, via
lab-provision-browserrot.sh / TestIntegration_realBrowserRotation);
real-site selector tables stay UNPROVEN and nothing auto-registers
into production rotate yet. 14 packages, -race clean.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Move the vision prose out of the centered header block (centered
multi-line text renders poorly), restructure into a numbered
two-parts list with labeled start/end-goal, fix typos.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
gopass writes GPG-encrypted blobs to disk and the backup gate writes sealed
.age bundles; only *decrypted* secrets are RAM-only. Fix the tagline, ASCII
box, banner alt-text, and banner.svg prose to state the defensible claim.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- untrack CLAUDE.md (tailnet leak) + stale HANDOFF.md; keep local via .gitignore
- add source-available LICENSE (attribution on fork, royalty on commercial use)
- add lab/ reproduction kit (fake-cred LIVE-VM/mock rotation POCs) + lab/README
- rewrite README to current status (22 drivers, 8 LIVE-VM/14 MOCK-ONLY/0 LIVE-REAL,
247 tests) and carry the credential-handling safety rules
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add ProofLiveReal (highest level, no driver yet) and enforce it at
execute time: rotate --execute only runs drivers with proof >= LIVE-VM
(MinExecuteProof). MOCK-ONLY/UNPROVEN drivers are skipped before any
provider call unless --allow-mock-proven is passed. ExecuteResult gains
Skipped/SkipReason; the CLI reports and audit-logs skips separately from
failures. Closes the last structural gap where an emulator-only driver
could touch a real credential unattended. Covered by execute_gate_test.go.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
policy 66% -> 94.7%: TestLoad exercises the previously-untested Load path —
empty/missing path fall back to Default, valid YAML parses onto the Default base
(absent fields keep their default; overrides apply), malformed YAML and a
non-regular file (directory) both error. Documents that a map override entry is
replaced wholesale by yaml.v3 (Evaluate re-merges defaults, so it's harmless).
tui 57% -> 87.8%: cover Init, the "o" open branches (URL vs none), "s" skip and
prev/next boundary guards, non-key messages, enter-at-last-item, esc/ctrl+c
quit, openURL's command, and Run's headless empty-list path. The residual is the
TTY-only tea.Program.Run() launch, not unit-testable headless.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
sink 30%->85%, cmd 4.3%->34.5%, gitea CI + Makefile landed; remaining coverage
gaps are policy (66%) and tui (57%). Keeps the L1 router honest for future
sessions (same anti-staleness rationale as M0).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Makefile targets per ROADMAP M1: build / vet / test / race / cover / check-cover
/ map / vm-proof. check-cover reports any package below a 70% floor (advisory —
not yet a hard gate, since rotate/policy/tui are still below target).
.gitea/workflows/ci.yml enforces the non-negotiable green bar on push/PR:
build + vet + `go test -race ./...` on Go 1.25, installing bash+openssl for the
self-contained test suite. Coverage floors are reported but not enforced until
the lagging packages are lifted (ROADMAP M1). Remote is gitea only.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
cmd/incredigo 4.3% -> 34.5%. Extract newRootCmd() as a testability seam so tests
drive the real wired CLI (persistent flags + every subcommand) via SetArgs.
main_test.go: a runCLI harness that captures os.Stdout/os.Stderr around Execute,
plus a fake gopass on PATH + fixture HOME so no command touches the user's real
store or host dotfiles (discovery isolated to --source file). Covers the safety-
critical contracts: rotate --execute refused without INCREDIGO_ALLOW_EXECUTE=1;
backup gate aborts BEFORE any driver runs when the snapshot fails; empty
passphrase rejected with no sealed content left behind; export refuses to
overwrite; unknown sealer rejected. Plus scan/status/worklist happy paths,
migrate --dedupe, and an export -> import round-trip proving byte-exact restore
with no plaintext in the sealed bundle.
discover: add ClearPaths() to reset the file scanner's process-global --path
targets between CLI invocations (a test-binary need; the one-shot CLI never hits
the stale-target accumulation).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The backup gate is only as trustworthy as the sealers; internal/sink was the
safety-critical coverage gap blocking all real rotation work.
sealer_test.go: round-trip per sealer (age/hmac/openssl) across ascii/binary/
unicode/empty/large payloads, asserting plaintext never leaks into sealed
output; tamper fails closed (age & hmac error with zero bytes emitted on a
flipped ciphertext or MAC byte); openssl's unauthenticated failure mode pinned
as an executable contract (no exact-plaintext recovery only); wrong-passphrase
clean-error; short-bundle, bad-binary, and empty-passphrase error paths.
framing_test.go: bundle framing (zero-record, clean-EOF, truncated mid-path/
len/secret, oversized path); gopass wrapper Insert/InsertAt/Show/Exists/
Available; secret-never-in-argv leak assertion (Hard Rule #3); gopass-failure
paths surface errors instead of silently short backups; end-to-end export ->
seal -> open -> count -> restore proving completeness + authenticity together.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Record the 2026-07-14 development snapshot and the roadmap to a deployable
rotation system. Update CLAUDE.md "Where development is" — the stale entry
claimed zero rotation drivers; reality is 22 drivers + a wired Mode A execute
spine (8 LIVE-VM / 14 MOCK-ONLY). Gitignore the incredigo-map.html build
artifact and the local rotation runbook.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
GitHub PATs and Stripe secret keys have no scriptable rotation API (GitHub's
create-token API was removed in 2020; Stripe has no create-key endpoint), so
they belong in the guided change-password layer, not as Rotators. But they
arrive from the env/file scanner as generic Source="env" tokens with no host,
so the worklist showed them as "manual — no web page".
Recognise them by their well-known PUBLIC value prefixes (ghp_/gho_/ghs_/
github_pat_/…, sk_live_/sk_test_/rk_live_/…) at scan time and record a
non-secret Meta["service"] hint — a fixed service NAME, never the secret bytes.
discover.ServiceForSecret does the detection; env.go attaches it for generic
tokens, file.go before Store wipes the buffer. links.HostFor consults the hint
and maps github→github.com / stripe→stripe.com, so the curated change-password
URLs (already in the table) now light up for these credentials.
Leak-safe (service name is derived from a public prefix, not the secret) and
verified by the existing assertNoLeak checks. go build/vet clean; full suite
179 tests pass.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Two more provider-API rotation drivers for the vibe-coder credential surface,
both following the create-new → verify → revoke-old pattern with the
self-contained blob form <provider>://<host>/?id=&secret=.
- vercel: POST /v3/user/tokens mints a new bearer; Verify lists tokens and
asserts the new id is present; RevokeOld DELETEs the old token by id.
- sendgrid: clones the old key's scopes (GET /v3/api_keys/<id>) onto the new
key so rotation is a faithful replacement, not a privilege downgrade; Verify
is a GET /v3/scopes auth proof; RevokeOld DELETEs the old key by id.
Both ship Bearer-enforcing httptest emulators that prove a real cutover (old
credential dead, new alive after RevokeOld) plus a leak check. Recorded as
MOCK-ONLY in proofs.go + docs/ROTATION-PROOFS.md (no self-host). go build/vet
clean; 77 rotate tests pass, -race clean.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
First of the vibe-coder API-rotation batch. Resend mints a new API key
authenticated by the old one (POST /api-keys), verifies it by listing keys
with the new Bearer token and asserting its id is present, then deletes the
old key by id (DELETE /api-keys/<id>). Self-contained blob
resend://host/?id=&secret= keeps the token off Identity/Meta/logs/argv.
httptest emulator enforces Bearer validity (key authenticates only while it
exists) so the test proves a real cutover: old dead, new alive, plus a leak
check. Recorded MOCK-ONLY in proofs.go + ROTATION-PROOFS.md (no self-host).
Note: GitHub PAT is intentionally NOT an API Rotator — GitHub has no
create-token API (classic Authorizations API removed 2020; fine-grained PATs
are UI-only), so a PAT belongs in the guided/worklist path, not here.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Group 2 of the manager wire-up — two CLI ItemUpdaters:
pass/gopass: the user's own password-store as a rotation source. Reads
line-1=password + key:value metadata via `<bin> show`; updates in place by
piping the full body to `<bin> insert -m -f` (off-argv), preserving metadata
across the password swap. gopass enumerates via `ls --flat`, pass by walking
the store dir; --pass-prefix scopes a subtree.
keychain: macOS internet passwords via the `security` CLI. Pure-exec (no
build tag) so it unit-tests cross-platform via a fake bin; Available() is
false off darwin. Read off-argv (find-internet-password -w); write is
delete+add with -w on argv (CLI limitation, documented like 1password).
Both MOCK-ONLY (fake-binary unit tests + leak checks); recorded as DATA in
BROWSER-ROTATION.md §8. 153 tests green, vet clean.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
One data-driven csvManager (layout = header + record func) covers
edge/brave (chrome layout), safari, lastpass, dashlane, protonpass,
nordpass, roboform, and read/guide-only enpass — replacing 8 near-identical
files. Export is header-name driven (parseLoginCSV + new aliases); import
stages a per-manager-layout CSV on tmpfs and shreds it, gated by --allow-csv.
Notes now round-trip through Meta. Reimporter exposes per-manager re-import
hints. Column layouts are MOCK-ONLY (unit-validated) on the same LIVE-VM
staging machinery; recorded as DATA in BROWSER-ROTATION.md §8.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Proven end to end (scan → plan → commit → restore-from-backup) against real
bw 2026.5.0 driving a self-hosted Vaultwarden 1.36; fake account minted via
real registration crypto. No adapter code changed. Only 1Password remains
MOCK-ONLY (needs a real account, no self-host). Proof recorded as DATA in §8.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
lab-provision-browsercsv.sh + csv-commit-probe.py drive the real `passwords
commit --allow-csv` path in the sandbox VM and assert the security-critical,
unattended machinery: the new-password CSV is written to a real tmpfs
(/dev/shm), in the correct per-browser column layout, carrying the staged
strong passwords with the OLD ones absent, then securely shredded once the
human confirms — and the user's own export CSV is left untouched. The final
re-import into the browser remains a human step (no programmatic import API),
so the proof is scoped to staging, not end-to-end browser write.
No adapter code changed; only the validation-status DATA in §8.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Proven end-to-end (scan→plan→commit→restore-from-backup) against the real
keepassxc-cli 2.7.6 in the incredigo-sbx sandbox VM. The live run surfaced an
adapter bug the fake CSV masked: KeePassXC's CSV "Group" column is the full path
including the root group name (default root is "Passwords"), but keepassxc-cli
addresses entries relative to root (/GitHub, not /Passwords/GitHub). kpEntryPath
now drops the leading root segment; the unit test that encoded the buggy
/Root/GitHub form is corrected and a nested-group table test added.
Records the proof as data: docs/BROWSER-ROTATION.md promotes keepassxc
MOCK-ONLY -> LIVE-VM (the others stay MOCK-ONLY).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add the secondary-persona path from docs/BROWSER-ROTATION.md: ingest a password
manager / browser store, take a mandatory verified sealed backup, generate fresh
strong passwords, hand the human a ready-to-finish task at the MFA wall, and commit
the new value into the manager only after the site change is confirmed.
- internal/pwgen: crypto/rand generator (rejection sampling, per-class guarantee,
vault-native — never returns plaintext as a Go string).
- internal/pwstore: Manager/ItemUpdater/BulkImporter contracts, secrets-free
RedactIdentity, and five adapters — bitwarden (bw, stdin), keepassxc
(keepassxc-cli, stdin), 1password (op, argv assignment with documented caveat),
chrome/firefox (tmpfs CSV ingest-and-shred). All real code; validation status is
data (MOCK-ONLY against fake binaries/CSVs, recorded in the design doc).
- internal/pwstore/stage.go: age-sealed staged-list (WriteStaged/ReadStaged) +
StagedImporter — the only artifacts crossing the plan->commit gap, never plaintext.
- cmd/incredigo: passwords scan|plan|guide|commit. Backup gate seals + round-trip
verifies before any commit; guide is interactive verify-before-commit; commit is
headless over a sealed stage. Browser CSV writes are gated behind --allow-csv.
Hard rules honored: backup-before-commit, verify-before-commit, no plaintext on
disk (browser CSV is the flag-gated tmpfs+shred exception), MFA always a human
handoff. go vet + -race clean; end-to-end verified against a fake bw.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Ran live cutover POCs in the sandbox VM against the real target software and
promoted both drivers from MOCK-ONLY to LIVE-VM in the proof table (data, not
driver behaviour):
- k8s: proven against real k3s v1.35 kube-apiserver + token controller
(lab-provision-k8s.sh). The real apiserver serves a self-signed cert, so the
driver gained CA-pinned TLS — blob params ca= (base64 PEM, pinned as the only
trusted root) and insecure=1 (lab-only escape hatch), routed through a new
clientFor(cr). Neither is test-awareness; a real deployment configures the same
CA. Added TestK8sTLSTrust covering CA-pinned / insecure / untrusted-rejection
and the blob round-trip.
- mongo: proven against real mongod 8.0 (lab-provision-mongo.sh).
npm stays MOCK-ONLY by decision (registry.npmjs.org not self-hostable; real
create-token requires the account password in the body) — documented as a
contract gap rather than faking a LIVE-VM. gcp/twilio/flyio remain MOCK-ONLY.
Live POCs surfaced real-target-only behaviour now recorded in the docs: the
apiserver's ~10s successful-auth cache (old token kept working ~7s after Secret
deletion) and real mongosh's stdout prompt. Full suite 104 green, -race clean.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Each driver is real-service code only (no test awareness; tests inject just an
HTTPClient/Bin), following the established patterns:
- mongo: in-place changeUserPassword over the old connection; no secret on argv
(host/db on the URI, both passwords on stdin, output scrubbed).
- npm: create token -> /-/whoami -> delete-by-key; blob records the key.
- gcp: service-account KEY rotation; mints an RS256 JWT signed with the SA's
own private key (crypto/rsa, stdlib) to self-auth the IAM create/verify/delete.
- twilio: API Key create -> get -> delete, Basic auth (KeySid/secret).
- flyio: GraphQL — a managing token mints/deletes the rotated deploy token.
- k8s: create a service-account-token Secret, await the populated token,
verify against the SA, delete the old Secret (invalidates the old token).
All 6 register their honest proof level (MOCK-ONLY) in proofs.go +
ROTATION-PROOFS.md; mongo/npm/k8s are LIVE-VM candidates pending real-target VM
POCs. Emulators enforce credential validity (gcp verifies the JWT signature),
tests assert Verify(old) fails after revoke + a secret-leak canary.
Full suite: 103 green, -race clean on rotate/sink/vault.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add gitlab, cloudflare, ghactions (MOCK-ONLY) and appsecret (LIVE-VM) one-file
Rotators, each with a table-driven cutover-proof + leak-check test. gitlab/
cloudflare/ghactions are in-place SaaS-token rolls (self/rotate, value-roll,
sealed-secret overwrite) so RevokeOld is a no-op; ghactions seals via
nacl/box.SealAnonymous (no new go.mod dep). appsecret regenerates a local app
signing secret and rewrites it in place across every target file (atomic, mode-
preserving, redacted errors), discoverable via exact-match app-signing key names
in env.go.
Keep real-rotation code distinct from mock code: how each driver's cutover was
validated lives as DATA in internal/rotate/proofs.go (single source of truth) +
docs/ROTATION-PROOFS.md, surfaced as a PROOF column in `incredigo rotate` so a
MOCK-ONLY driver can never be mistaken for a LIVE-VM one. appsecret proven LIVE-VM
against real local files in the sandbox VM.
82 tests green, -race clean on rotate/sink/vault.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Adds VISION.md (target personas + 20-credential surface), docs/MARKET.md
(competitive positioning), and docs/FINDINGS-2026-06-16.md (controlled-env
test run: live cutover POCs for postgres/gitea/aws/mysql/redis/wireguard, the
backup-gate defect + fix, and the MariaDB unprivileged-rotation gap). Swaps the
placeholder _igo_options assets for the banner/icon set. Refreshes the project
CLAUDE.md directive.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
internal/blast builds a read-only consumer map (which other files appear to
use each credential) from NON-secret metadata only — env-var names, hostnames,
identifiers — never the vault secret; rotation needs it to know what to
redeploy after a change. cmd/incredigo wires Mode A `rotate --execute`
(reconstructs Source from the gopass path segment, runs the execute spine) and
surfaces the blast-radius map. Adjusts a worklist test accordingly.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
An env credential's Identity (".env / API_TOKEN") produced StorePath
imported/env/.env/api_token; gopass hides dot-prefixed segments from
`ls --flat`, so those entries were invisible to bundle.ListPaths and the
mandatory backup gate silently under-covered them (Hard Rule #1 violation).
slug() now strips leading dots per /-separated segment so incredigo can no
longer write a path that `ls --flat` hides. Locked in by
TestStorePathNoDotSegment.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Per the standing directive (dynamic discovery per credential type as each
driver lands):
- env.go: tag DB connection URLs by scheme (postgres/postgresql, mysql/mariadb,
redis/rediss) with a user:password authority, routing DATABASE_URL/REDIS_URL/
MYSQL_URL straight to the in-place DB drivers, kept regardless of the
secret-name heuristic.
- aws.go: emit the driver-ready single-line blob (was the bare secret) and
resolve region from env / ~/.aws/config.
- ssh.go: content-based key detection (not just id_*) + ~/.ssh/config
IdentityFile parsing.
- tea.go: new scanner mapping Gitea `tea` CLI config to a driver-ready PAT blob.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Implements the Rotator interface across all four rotation patterns, each a
self-contained one-file driver self-registering via init():
in-place DB password postgres, mysql (3-stmt unprivileged fallback), redis
local keypair + propagate sshkey (ed25519), wireguard (clamped curve25519)
provider-API token gitea PAT, mullvad device key
cloud-key self-identifying aws IAM access key (hand-rolled SigV4, no SDK dep)
Spine (execute.go) enforces Hard Rule #2: backup -> rotate -> verify(new) ->
store -> re-read+verify -> revoke-old; dryrun.go keeps --execute gated. Each
driver ships a table-driven test proving real cutover (old secret stops
authenticating) and asserting no secret substring leaks to argv/Meta/errors.
Promotes golang.org/x/crypto to a direct dependency.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Operational directive for Claude sessions: the headless two-mode rotation
goal, current phase, and the non-negotiable guardrails (backup-before-rotate,
verify-before-revoke, no-plaintext-on-disk / browser worklist not dump,
GOPASS_HOMEDIR test isolation, self-owned-accounts scope, PGP passphrase =
root of trust). Chosen stack: Bubble Tea + go-rod + .well-known/change-password.
docs/ROTATION.md Part II (capability matrix, TUI walkthrough, link gen, browser
worklist, automation tiers) is updated on disk but stays gitignored (local-only).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Design-phase foundation for v2 rotation. Changes NO credential at any service.
- internal/rotate: Rotator interface + registry + PlanAll (zero drivers
registered, so every credential plans as "(none)")
- internal/rotate.Snapshot: the MANDATORY backup gate — seals the gopass
prefix into an authenticated bundle, re-opens it, and confirms the entry
count matches before returning; any failure blocks rotation
- internal/sink.CountBundleRecords: read-only bundle completeness check
- cmd: `incredigo rotate` runs the backup gate + prints the plan; it is
dry-run only and refuses --execute (design phase)
- tests: backup gate happy path, no-overwrite, tamper detection; race-clean
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
leetcrypt
Trilltechnician