TESTING.md

Testing Guide

Run the right ssh-cli test profile without hanging on remote networks.

• Read this document in Portuguese (pt-BR)
• Product line: 0.5.6 (historical residual suite gates include 0.4.1 AUD-POST / gaps_v041 and 0.5.2 wire / gaps_v051)

Why Categorized Tests

• Unit tests protect packing, schema, secrets authenticated encryption (AEAD), and pure logic without SSH servers
• Integration tests protect command-line interface (CLI) contracts, storage, and snapshots
• Optional ssh-keygen fixtures (G-PROC-02) generate real OpenSSH keys for key-path tests
• Missing binary skips those cases: product runtime never spawns it
• Remote live tests are optional and must always use hard timeouts and never log credentials
• Install resolve gates protect crates.io onboarding (GAP-014)
• Residual gap suites lock agent I/O, exit codes, supply chain, masking, Secure Copy (SCP) and SSH File Transfer Protocol (SFTP) wire, and doc honesty contracts
• The audit record is a local maintainer file, gitignored, cargo-excluded and never published, and tests must not assert its FIXED text (G13/G15)
• G6: tests that touch signal/cancel global state (CANCEL_FLAG) use #[serial_test::serial] (dev-dep serial_test) so the suite is deterministic
• Do not remove serial markers from concurrency/signal tests
• G11: baseline must stay green on first run
• Re-running until pass is forbidden as a gate strategy

Test Categories

• Unit tests inside the src modules (includes secrets default encryption)
• Unit tests of ssh-cli vps edit in src/vps/edit/apply_tests.rs use tempfile::tempdir(), never a fixed path in the system temp folder, so parallel runs cannot collide
• CLI e2e under tests/e2e_cli.rs
• Gap/residual integration under tests/gaps_v035_integration.rs (fake secrets only)
• Agent I/O residual suite under tests/gaps_v037_integration.rs
• Post-0.3.7 residual suite under tests/gaps_v038_integration.rs
• Post-0.3.8 residual suite under tests/gaps_v039_integration.rs (LOG-001, JSON-001, CLI-004, DOC/DENY/CHG)
• Post-0.3.9 / 0.4.0 suite tests/gaps_v040_integration.rs
• AUD-POST suite tests/gaps_v041_integration.rs (EXP-001, TUN-002, CLI-005/006, IO-009, REL-006, DOC-041)
• AUD-E2E suite tests/gaps_v042_integration.rs (TUN-003, IO-010, UX-001, REL-007, ENV-001, DOC-042, SCP-024)
• 0.5.2: suite tests/gaps_v051_integration.rs (export redaction roundtrip, vps-export JSON, schema v3 dual-read, secrets-init event, include-secrets guard, CRUD vps-added, empty command, import exit 65)
• Domain types suite tests/gaps_v053_domain_types.rs
• Error handling suite tests/gaps_v054_error_handling.rs
• Unsafe/FFI residual suite tests/gaps_v055_unsafe_ffi.rs
• G-SFTP residual suite tests/gaps_v057_sftp.rs (SFTP surface, with destination-effect proof preferred over inventory self-cert)
• G-E2E residual suite tests/gaps_v058_e2e_residual.rs (root ssh-cli schema / ssh-cli doctor, single vps-added with secrets_key_auto_created, --use-agent, help/clap env purge, ambient RUST_LOG ignored, ACME exit 64, etc.)
• Storage integration under tests/storage_integration.rs
• Snapshot tests under tests/snapshot_tests.rs
• SCP surface under tests/scp_integration.rs
• Tunnel surface under tests/tunnel_integration.rs
• Tunnel mode surface under tests/gaps_v060_tunnel_modes.rs: the 0.5.4 modes --reverse, --socks5 and --remote-socket, their mutual exclusion, the mode wire label (local / reverse / socks5 / streamlocal), the --i-accept-network-exposure guard on both ends, and the tunnel_closed event
• Agent-native output surface under tests/gaps_v059_agent_native.rs: --select, --filter, --limit, --count-only and the truncation report
• Error taxonomy under tests/gaps_v061_error_taxonomy.rs: error_class, retryable and suggestion on every envelope
• SCP durability under tests/gaps_v061_scp_durability.rs: mtime_preserved and durable are reported, never assumed
• Component budget under tests/gaps_v062_component_budget.rs: the ratchet fails both on growth past a frozen cap and on a stale entry whose file already shrank, so the ledger records repayment as loudly as debt
• i18n reachability under tests/gaps_v062_i18n_reachability.rs: every message key is reachable in both languages
• Gate falsification for scripts/check_en_identifiers.py, driven by scripts/check_all_gates.py: each check must be provably able to fail, and two of them were dead from introduction
• The standing proof now lives in tests/gaps_v068_source_language.rs
• E2E harness, locale features and secret stdin under tests/gaps_v063_e2e_harness.rs, tests/gaps_v063_locale_features.rs, tests/gaps_v063_secret_stdin.rs
• Explicit target designation under tests/gaps_v065_exec_target_designation.rs: the argv of GAP-SSH-EXEC-ARGC-001 stays rejected, and the provenance trio reaches success and error envelopes alike
• Transfer target designation under tests/gaps_v066_transfer_target_designation.rs: the same rule on scp and sftp
• Packaging reach under tests/gaps_v067_packaging_reach.rs: every source file survives cargo package
• A .gitignore pattern that swallows a source directory is invisible to every local build
• Source language under tests/gaps_v068_source_language.rs: code and comments are English, exempting citations rather than whole files
• Markdown table rendering under tests/gaps_v069_markdown_tables.rs: a row whose unescaped pipes disagree with its delimiter row is valid Markdown that renders with phantom columns
• Documentation reference reach under tests/gaps_v070_doc_reference_reach.rs: every repository path a document names exists on disk, and every suite on disk is catalogued here
• The phantom suite this closed was invisible to the forward sweep
• Edit and rename under tests/gaps_v071_vps_edit_rename.rs: the secret stays readable under the new name while the file keeps only ciphertext, the ssh-cli connect marker follows only when it named the source, neither verb retargets from the marker, the eight clear flags empty exactly their field and fail closed when they would leave zero auth, --dry-run writes nothing on either verb, and the config lock exits 69 instead of hanging
• Edit truthfulness under tests/gaps_v072_edit_truthfulness.rs, ssh-cli vps edit reports only what changed: a flag whose value already matches leaves updated empty (and the key stays in the envelope), a no-op exits 0 and leaves config.toml byte-identical, and a rename fails in a fixed order (unknown host 66, identical name 64, taken name 64)
• Import round trip under tests/gaps_v072_import_roundtrip.rs: a host exported with its secrets and imported into an empty registry compares equal field by field through JSON and TOML, Transport Layer Security (TLS) fields are validated on import, the replaced field is reported, a closed stdout has its own exit code, and a pinned field count fails the moment VpsRecord grows without export and import learning the new field
• Lock contention under tests/gaps_v072_lock_contention.rs: exit 69 on a held registry lock carries the lock remedy and not the keyring one, proven with a real second process holding config.toml.lock
• Schema conformance under tests/gaps_v072_schema_conformance.rs: real envelopes printed by the binary validate against the published schemas in docs/schemas/, through a small hand-written validator so no new dependency enters the supply chain
• Storage hardening under tests/gaps_v072_storage_hardening.rs: the lock and marker files keep their expected modes on disk after a real invocation, asserted on Unix only because Windows has no POSIX mode bits
• Auth method switch under tests/gaps_v073_vps_auth_switch.rs, ssh-cli vps edit switches a host from one authentication method to another in one call: --key clears the password and the agent, --password and --password-stdin clear the key, its passphrase and the agent, and --use-agent clears the password and the key
• The same suite proves that two primary auth flags in one call exit 2 with an empty stdout, a passphrase with no key exits 64 and leaves the file untouched, a different key drops the old passphrase, --agent-socket alone switches to the agent, and the --dry-run plans echo target_resolved and target_source
• Host name charset under tests/gaps_v073_host_name_charset.rs: a new host name is limited to ASCII letters, digits, dot, hyphen and underscore on ssh-cli vps add, ssh-cli vps rename, ssh-cli vps edit --name and ssh-cli vps import
• The same suite proves that ssh-cli vps import lists every bad name at once and writes nothing, legacy names with spaces or accents still load, and ssh-cli vps rename repairs a legacy name with control characters
• Lookups that become a target (ssh-cli exec, ssh-cli health-check, --hosts, ssh-cli connect) refuse control characters with exit 64, and ssh-cli tls mtls import --name refuses control, formatting and bidi characters without creating a directory
• Terminal escape under tests/gaps_v073_terminal_escape.rs: control and bidi characters are escaped as \u{1b}-style text in text mode (ssh-cli vps list, ssh-cli vps show, not-found errors and lookups) before they reach the terminal, while JSON keeps the original value
• Config directory mode under tests/gaps_v073_config_dir_mode.rs: the config directory is created with mode 0700, an existing default XDG directory is tightened, and any other existing directory is never chmodded while ssh-cli doctor reports config_dir_exposed
• The same suite proves that a path ending in .toml makes the parent the config directory, ssh-cli vps export creates missing folders private, two tables that normalize to one name exit 65, and a source scan fails when another module creates the directory
• Flag reach under tests/gaps_v074_flag_reach.rs: each documented flag is proven by observable behaviour instead of parser acceptance, and each document that claims the full inventory fails when it loses a flag
• Markdown format gate under tests/gaps_v075_markdown_format_gate.rs: scripts/check_markdown_format.py stays in the battery, runs against the real tree and imports only the standard library
• Documentation conformance under tests/docs_conformance.rs: the 48-leaf command inventory, schema indexing in both languages, bilingual release parity, the eight --dry-run operations, directional permission masks and the 7000-word skill budget
• Test quality under tests/test_quality.rs
• Property tests under tests/proptest_tests.rs
• i18n integration under tests/i18n_integration.rs
• Gate battery runner scripts/check_all_gates.py: runs all fourteen mandatory gates in one invocation
• The suites above run under cargo test, which the battery invokes as its test gate
• Advisory freshness gate scripts/check_advisory_freshness.py: only reachable through the battery runner
• Advisory hits gate scripts/check_advisory_hits.py: runs cargo audit --json --no-fetch and fails on any advisory against Cargo.lock, because a fresh database and a flat cargo deny still never stated which advisories hit the tree
• Gate battery coverage contract tests/gaps_v064_gate_runner.rs: every public scripts/*.py is a gate or a declared non-gate, and the_test_gate_stays_locked fails when the test gate loses --locked or --no-fail-fast
• Install resolve script scripts/verify_install_resolve.py
• English identifier gate scripts/check_en_identifiers.py
• Real SSH E2E (optional, machine-local): scripts/e2e_real_ssh.py: official matrix E01–E21 (E10–E14 cover SCP upload/download/cmp/missing/preserve, E17/E18 cover SFTP checksum + recursive tree, G7, and E19–E21 cover the reduced vps list --select item, the --limit accounting, and the keyring round-trip)
• Benchmarks under benches/ (manual)

How to Run

Full battery (required before declaring a gate green)

• One invocation runs every mandatory gate and reports every result

python3 scripts/check_all_gates.py

• Add --json for NDJSON, --only ID[,ID...] for a subset, --list for the gate ids
• The fourteen gates are fmt, build-release, build-no-default, clippy, doc, test, exec-target, deny, cross-targets, advisory-freshness, advisory-hits, en-identifiers, markdown-format and install-resolve
• The test gate runs cargo test --locked --all-features --no-fail-fast
• --no-fail-fast makes a red run show every failing test instead of stopping at the first
• Other flags are --fail-fast to stop at the first red gate, --log-dir DIR to keep the per-gate logs and --timeout-secs N for the per-gate budget (default 3600, 0 for none)
• --quiet hides the progress lines on stderr, and the tail of a failed gate still prints
• The stdout has one tab-separated line per gate, with the id, pass or fail, the exit code and the milliseconds, then a _summary line with total=, failed= and skipped=
• The exit is 0 when every selected gate passes, 1 when any gate fails and 2 on a usage error such as an unknown gate id
• The exec-target gate runs only tests/gaps_v065_exec_target_designation.rs, so a regression of the dangerous exec argv is named in the report
• cargo clippy and cargo test abort on the first unbuildable target, so a single broken test file hides every gate behind it
• The run always reports how many gates it skipped, so a partial run cannot read as a full one
• The battery is sequential on purpose: the cargo gates contend for one target/ lock, so running them concurrently buys nothing

Local developer loop

cargo fmt --check
cargo test --locked --all-features --no-fail-fast
cargo clippy --all-targets --all-features --locked -- -D warnings
python3 scripts/check_en_identifiers.py
cargo build --release
python3 scripts/verify_install_resolve.py

Focused profiles

• Every suite in tests/ runs alone with cargo test --locked --test <suite>
• The block lists all of them

cargo test --locked --test docs_conformance
cargo test --locked --test e2e_cli
cargo test --locked --test gaps_v035_integration
cargo test --locked --test gaps_v037_integration
cargo test --locked --test gaps_v038_integration
cargo test --locked --test gaps_v039_integration
cargo test --locked --test gaps_v040_integration
cargo test --locked --test gaps_v041_integration
cargo test --locked --test gaps_v042_integration
cargo test --locked --test gaps_v051_integration
cargo test --locked --test gaps_v053_domain_types
cargo test --locked --test gaps_v054_error_handling
cargo test --locked --test gaps_v055_unsafe_ffi
cargo test --locked --test gaps_v057_sftp
cargo test --locked --test gaps_v058_e2e_residual
cargo test --locked --test gaps_v059_agent_native
cargo test --locked --test gaps_v060_tunnel_modes
cargo test --locked --test gaps_v061_error_taxonomy
cargo test --locked --test gaps_v061_scp_durability
cargo test --locked --test gaps_v062_component_budget
cargo test --locked --test gaps_v062_i18n_reachability
cargo test --locked --test gaps_v063_e2e_harness
cargo test --locked --test gaps_v063_locale_features
cargo test --locked --test gaps_v063_secret_stdin
cargo test --locked --test gaps_v064_gate_runner
cargo test --locked --test gaps_v065_exec_target_designation
cargo test --locked --test gaps_v066_transfer_target_designation
cargo test --locked --test gaps_v067_packaging_reach
cargo test --locked --test gaps_v068_source_language
cargo test --locked --test gaps_v069_markdown_tables
cargo test --locked --test gaps_v070_doc_reference_reach
cargo test --locked --test gaps_v071_vps_edit_rename
cargo test --locked --test gaps_v072_edit_truthfulness
cargo test --locked --test gaps_v072_import_roundtrip
cargo test --locked --test gaps_v072_lock_contention
cargo test --locked --test gaps_v072_schema_conformance
cargo test --locked --test gaps_v072_storage_hardening
cargo test --locked --test gaps_v073_config_dir_mode
cargo test --locked --test gaps_v073_host_name_charset
cargo test --locked --test gaps_v073_terminal_escape
cargo test --locked --test gaps_v073_vps_auth_switch
cargo test --locked --test gaps_v074_flag_reach
cargo test --locked --test gaps_v075_markdown_format_gate
cargo test --locked --test i18n_integration
cargo test --locked --test proptest_tests
cargo test --locked --test scp_integration
cargo test --locked --test snapshot_tests
cargo test --locked --test storage_integration
cargo test --locked --test test_quality
cargo test --locked --test tunnel_integration
cargo test --locked --lib
cargo test --locked packing
cargo test --locked secrets::
cargo fmt --check

Documentation profile

• Run these suites after any change to a document, a schema or a skill
• They cover command inventory, schema indexing, Markdown tables, repository paths, suite catalog, packaging reach, flag reach, Markdown format and the gate runner

cargo test --locked --test docs_conformance --test gaps_v070_doc_reference_reach --test gaps_v072_schema_conformance --test gaps_v039_integration --test gaps_v064_gate_runner --test gaps_v067_packaging_reach --test gaps_v069_markdown_tables --test gaps_v068_source_language --test gaps_v074_flag_reach --test gaps_v075_markdown_format_gate

Real SSH E2E (never print secrets) — G-E2E-05

# Preferred: unprivileged sshd on loopback, nothing to configure (Unix only)
python3 scripts/e2e_real_ssh.py --local-sshd

# Lab host: flags only, no environment variables; the password comes from a file or stdin
python3 scripts/e2e_real_ssh.py --host … --port 22 --user … --password-file /path/to/pw-file --sudo-password-file /path/to/sudo-pw-file

# Maintainer-local only: parse $HOME/.grok/config.toml
# That file must stay under $HOME: never copy it into this repository. Needs Python 3.11+.
python3 scripts/e2e_real_ssh.py --from-grok-config

• Default binary: target/release/ssh-cli (override with --bin)
• On Windows use py -3 scripts/e2e_real_ssh.py ...
• --local-sshd is Unix-only
• --from-grok-config [PATH] is optional and maintainer-only, so other contributors use --local-sshd or the lab flags
• Without a lab host / credentials, the script exits 0 with SKIP, which is offline-safe and not a red gate
• Official matrix E01–E21
• E10–E14 = SCP upload, download, integrity (cmp), missing remote, preserve mode+mtime (SCP-023)
• E17 = SFTP upload/download checksum
• E18 = SFTP recursive tree (G7)
• E19 = vps list --select returns a reduced item that satisfies vps-list.schema.json (D6)
• E20 = vps list --select --limit drops records, reports the count on stderr, and --quiet silences it (D7)
• E21 = secrets init --keyring round-trips through the OS store across two processes, and skips with SOFT when no store answers (Gap 50)
• Script prints only PASS/FAIL/SKIP labels: never host, user, or password
• Residual gate suite: cargo test --locked --test gaps_v058_e2e_residual
• GAP-014 / fail2ban policy: prefer local sshd or a throwaway VPS
• FORBIDDEN: auth-failure storms against production hosts (fail2ban bans)
• Production VPS e2e only with care, IP whitelist / ignoreip, and no intentional wrong passwords

Publish Gates

• Maintainers run the local developer loop before every publish
• Publish gates include package dry-run, install resolve verification, bilingual docs parity and the English identifier check (python3 scripts/check_en_identifiers.py)
• They also include the residual suites gaps_v040, gaps_v041, gaps_v042, gaps_v051, gaps_v057 and gaps_v058, plus cargo fmt --check (G10)
• The canonical loop is cargo test --locked --all-features --no-fail-fast, clippy --all-targets --all-features -- -D warnings and cargo build --release
• python3 scripts/check_all_gates.py runs all of them, plus the cross-target type-check, in one invocation

Environment Variables

• Use --config-dir on CLI invocations to isolate config during tests (product does not read SSH_CLI_HOME)
• --allow-plaintext-secrets opts out of default encryption for tests that assert plaintext TOML
• Without that opt-out, first secret write auto-creates secrets.key and encrypts fields
• Default tracing level is error
• Do not expect INFO prose on stderr by default
• Ambient RUST_LOG is ignored
• Use -v/-vv/-vvv when diagnosing failures (G2/G14 crate-scoped)
• -v → info, -vv → debug, -vvv → trace (CLI-only log filter)
• Use --no-color when snapshot-sensitive output needs stable text, because the standard color variables only drive terminal detection
• The product ignores SSH_CLI_HOME, SSH_CLI_LANG, SSH_CLI_FORCE_TEXT and SSH_CLI_USE_KEYRING as configuration, and it rejects SSH_CLI_SECRETS_KEY and SSH_CLI_SECRETS_KEY_FILE fail-closed
• Every command that reads or writes a registry secret exits 64 when either secrets variable is set, while ssh-cli vps path, ssh-cli locale show, ssh-cli schema and ssh-cli commands exit 0
• Never put live host passwords into env vars that tests print

Troubleshooting

• Snapshot drift: review tests/snapshots/ and update only intentional UI changes (including version strings)
• Crypto resolve failures: re-check pins and rerun the install script without ignoring lock policy
• Flaky timeout tests: ensure no real remote host is required unless explicitly configured
• Permission failures: confirm temp dirs are writable and mode assertions match the OS
• Encrypted fixture surprises: pass --allow-plaintext-secrets or provide a test primary-key via --secrets-key-file / XDG secrets.key
• Unexpected quiet stderr: default is error-level tracing
• Pass -v/-vv/-vvv if you need more lines (ambient RUST_LOG is ignored)
• Intermittent cancel/signal failures: confirm #[serial_test::serial] remains on tests that touch global cancel state (G6)
• Do not parallelize those cases
• Baseline red without code change: fix the suite, re-run lottery is not a gate (G11)
• Publish gates include cargo fmt --check (G10)
• SCP / AUD-POST / 0.5.2 / SFTP / G-E2E residual failures: run cargo test --locked --test gaps_v040_integration, gaps_v041_integration, gaps_v042_integration, gaps_v051_integration, gaps_v057_sftp, and gaps_v058_e2e_residual
• The local audit record may hold maintainer notes but is not a published gate artifact and tests must not assert its text (G13/G15)
• A pending *.snap.new file fails no_pending_snapshot_files_exist: accept the snapshot or delete the file
• A red documentation suite after a document edit: run the documentation profile above and read which path, schema or table row it names
• Never leave __pycache__ or .pyc files in the tree: run Python with python3 -B

Release Notes

0.5.6 — formatting gate

• scripts/check_markdown_format.py makes the formatting contract executable, closing what Gaps 25 and 37 named as their acceptance criterion while no such linter existed
• The battery runs fourteen gates, and markdown-format and doc are the two newest
• Thirteen rules cover bold, italic, horizontal rules, star bullets, trailing period or comma on a bullet, H4 and deeper, H1 count, H1 and H2 spacing, the header-to-bullet gap, em dashes outside an H3, decorative blockquotes and emoji
• --selftest carries one fixture per rule, so a regex that matches nothing cannot report a clean tree
• docs_rules/ and docs_prd/ stay out of scope as imported source material, declared rather than silently skipped
• tests/gaps_v075_markdown_format_gate.rs requires the gate to stay in the battery, to execute against the real tree over at least thirty documents, and to import nothing outside the standard library

0.5.6 — keyring and contract

• secrets init --keyring writes to the operating system store through keyring 4.2.0 with the v1 feature, and Entry::store_status() makes an unavailable store exit 69 instead of answering success
• The store entry is scoped by a digest of the resolved config directory, so --config-dir profiles no longer share one machine-wide key
• --use-keyring also reaches the init path, and the two flags can no longer disagree about the store
• The minimum Rust version is 1.89, required by russh 0.65
• tests/gaps_v074_flag_reach.rs proves each of the eight flags by observable behaviour instead of parser acceptance, and fails per document when one loses a flag from its published inventory
• The E2E matrix is E01–E21: E19 and E20 cover the reduced vps list payload and E21 the keyring round-trip
• E21 rotates a live profile into the store with --force, so it executes the transition an operator actually makes and fails when a leftover secrets.key would shadow the new key
• E20 provisions its own second registry entry, because a second host is what the case needs and vps list never dials, so a SOFT skip there was a named coverage hole, not a passing run
• secrets_init_refuses_to_fork_the_primary_key covers the defect E21 found: a second secrets init on a live profile must exit 64 and leave the key untouched
• docs/schemas/vps-list.schema.json declares both item shapes, so a --select item validates
• deny.toml carries three unmaintained ignores with written reasons, and the suite asserts the ignore set is exactly those three

0.5.6 — audit closure

• --select and --truncate-content reduce a single record, and a collection-only flag on one record is named under agent_shape.ignored even when it is the only flag passed, covered by src/agent_shape_tests.rs, tests/gaps_v059_agent_native.rs and tests/gaps_v072_schema_conformance.rs over both published shapes
• A reduced bare array writes {"event":"agent-shape",...} to stderr, covered by tests/gaps_v059_agent_native.rs
• --quiet suppresses that stderr line like every other one
• tests/gaps_v074_flag_reach.rs asserts behaviour for the 18 flags that had none and checks the documents that claim the full inventory
• tests/docs_conformance.rs requires an Examples: block on all 48 leaves and runs every published argv with --help appended
• scripts/check_advisory_hits.py fails the battery when an advisory reaches the lockfile, and the clippy gate now carries -W clippy::doc_markdown
• The battery runs fourteen gates
• The pure-Rust crypto migration was tried and reverted, so the TLS provider stays aws_lc_rs and the suites written for that migration were removed with it

0.5.6 — edit and rename

• ssh-cli vps rename <NAME> <NEW_NAME> and ssh-cli vps edit <NAME> --name <NEW> change a host name natively, with the credentials untouched
• The record seals each secret with an unbound context, so the associated data carries no host name and the secret stays readable under the new name
• Every save re-seals with a fresh nonce, which is normal and never touches the plaintext
• This replaces the export/edit/import detour, which silently dropped agent authentication because the export carried no agent fields
• Eight clear flags remove a field instead of overwriting it
• --dry-run covers both verbs
• The config lock waits a bounded --lock-wait-ms and then exits 69 instead of hanging
• ssh-cli commands --json publishes every argument

0.5.5 — target-designation security release

• Fixes GAP-SSH-EXEC-ARGC-001: ssh-cli exec/ssh-cli sudo-exec/ssh-cli su-exec decided the target host by positional count
• So ssh-cli exec <HOST> --step <CMD> executed the host name as a remote binary on whatever host ssh-cli connect last selected
• BREAKING: a lone positional is now a usage error
• Reach the active marker with the new --use-active <COMMAND>
• The single-host envelope gained target_resolved and target_source, so bind new consumers to those
• host_resolved and host_source are the 0.5.5 read aliases
• The envelope also gained active_fallback, and ssh-cli doctor now reports local.active_vps

0.5.4 — security and agent-native release

• Fixes a remote pre-auth DoS in the SSH banner path (A1)
• Stops server-sent setuid bits landing on downloaded files (A3)
• Closes the world-readable window on ACME/mTLS private keys (A2)
• Adds payload-shaping flags applied before serialization: --select, --filter, --limit, --sort, --dedupe-by, --count-only, --truncate-content, --max-output-bytes
• BREAKING: partial multi-host failure now exits 1 (was 65)
• BREAKING: a non-loopback --bind requires --i-accept-network-exposure
• New tunnel_closed event