Skip to content

Releases

icx v1.3.1

Choose a tag to compare

@github-actions github-actions released this 22 Nov 06:43
10e1f63

Fixes #28, #29 and #30. Release notes:
docs/releases/v1.3.1/.

Highlights

v1.3.1 hides a value after a sensitive key from its first quote to the end of the line, so secrets
inside code (env::var("K").unwrap_or("..."), x or "...", template and prefixed literals) no
longer stay visible in icx's stored copy (#29). A call head such as env::var( stays visible;
later text on the line is hidden with the value. Complete JSON lines, such as minified API output,
are parsed instead, so only the sensitive values are hidden. It also rejects CLI
flags with a missing or flag-shaped value (#28), and aligns hook fail-open replies with each
adapter's own no-op (#30). Details: v1.3.1 release notes.

Changed

  • Redaction (#29): after a sensitive key, a value that holds any quote is hidden from its first
    quote to the end of the line, as one marker. A call head right before that quote stays visible
    (password = env::var([REDACTED:ENV_SECRET]); any other value is hidden from its start, since a
    bare head could itself be the secret (hunter2 "note"). There is deliberately no quote
    pairing, no field boundary and no key detection: each was bypassed in review (a stray quote
    shifted the pairing; redacted key names hid later sensitive keys). So later text on the same
    line is hidden with the value, JSON keys and later CLI flags included. Pinned by a corpus of
    every reported shape and a property test over random code shapes.
  • Redaction (#29): after a sensitive key, no exemption (null default, type annotation, code
    expression, token counter) applies when a quote follows on the line.
  • Redaction (#29): a line that is one complete JSON object or array, such as minified API
    output, is redacted structurally by a strict RFC 8259 scanner. The value of a sensitive key, of
    any type, becomes "[REDACTED:CONFIG_SECRET]". Every other string is redacted on its own, with
    its key as context. Formatting and key order are kept, and the edited line must still parse.
    Anything that is not valid JSON, or nests deeper than 128 levels, keeps the first-quote rule.
    String values are decoded before redaction (UTF-16 surrogate pairs included), so escaped
    newlines and \uXXXX escapes cannot hide a key; embedded JSON and digest lines inside a string
    keep their own context, and a line with a complete inline key block stays on the JSON path.
  • Redaction (#29): a backtick inside a type annotation, or a quote in a YAML block header's
    comment (password: | # "..."), is hidden too. A
    7 KB minified API list with "token" first now keeps over 99% of its text instead of 0.5%.

Fixed

  • CLI (#28): every value-taking flag (--dir, --lines, --idle-exit, --max-requests,
    --listen, --provider, --upstream, --profile, --context-tokens, ...) rejects a missing,
    flag-shaped (--dir --raw, -h) or blank value instead of taking the next flag or falling back
    to a default. icx hook still fails open.
  • Hook fail-open replies (#30) come from one function in icx-adapters. They have the shape of
    each adapter's own no-op, pinned by a test against AntigravityAdapter: agy unknown events
    return {}, and agy Subagent events return injectSteps without terminationBehavior. Hermes
    gets {}. Blank --agent / --event values are rejected.
  • Muse (#30): a Bash result with an empty output falls back to stdout / stderr.

Documentation

  • Muse loads MCP servers only in a trusted workspace; hooks run either way (deployment guide §2.6,
    adapter spec §6).
  • README reworked as a user-facing landing page: a Getting Started section (verified release-binary
    install, per-agent setup links, first steps), a CLI Usage section, ADR-0006, the v1.2.0 audit
    and release notes in the documentation lists, and badges grouped in a titled table by topic.
    The per-release "What Changed" summaries and the phase-by-phase roadmap history moved into this
    changelog as "Highlights" under each version.
  • .gitignore: an in-repo workspace store (icx.sqlite3 and its WAL/SHM files), install.key,
    and local agent and environment files.
  • Release notes state that stored content is not re-checked after an upgrade.

Build

  • The Antigravity PR review pins AGY_PRINT_TIMEOUT to 15 minutes (a 79 KB diff timed out at
    the scaled 367 s); .gitignore already excludes Playwright MCP output (.playwright-mcp/).

Downloads

Linux x86-64 binaries in two C-library flavours:

  • *-linux-musl: fully static, runs on any Linux distribution.
  • *-linux-gnu: dynamically linked against glibc 2.35 or newer (Ubuntu 22.04,
    Debian 12, RHEL 10 and later); faster in-process hooks than musl.

The x86_64_v3 builds need AVX2 (Intel Haswell / AMD Zen or newer) and exit with
"illegal instruction" on older CPUs; use the baseline x86_64 builds there.
Verify with sha256sum -c SHA256SUMS --ignore-missing or
gh attestation verify <archive>.tar.gz --repo doublegate/Infinity-Context.
Third-party license notices: THIRD-PARTY-NOTICES.txt (also inside each archive).
Corresponding source for every binary is this tag.

icx v1.3.0

Choose a tag to compare

@github-actions github-actions released this 22 Nov 01:49
02daaaa

Adds Meta Muse Code as a fifth agent. Release notes:
docs/releases/v1.3.0/.

Added

  • Meta Muse Code adapter (icx hook --agent muse, AgentKind::Muse, MuseAdapter). Built
    from Muse Code 1.4.25's observed hook contract (probe hooks on a real session): Claude-Code-shaped
    payloads, additionalContext on SessionStart/UserPromptSubmit/PostToolUse, hookEventName
    required in hookSpecificOutput. SessionStart and UserPromptSubmit return the tail slot and
    recall; PostToolUse/PostToolUseFailure capture the output (the Bash tool's JSON result is
    unwrapped to its text and exit code). Muse rejects replacing tool output, so icx stores and
    recalls Muse output but does not shrink it. Every other event gets {}; no reply has a null.
    icx doctor detects the Muse binary and settings. Setup: deployment guide §2.6.
  • AgentKind::from_repr is now the single decoder of the agent byte on the telemetry bus, and a
    test round-trips every variant, so a new agent cannot be decoded as GenericMcp (which would
    echo Muse's own events back to it as sibling activity).

Fixed

  • Redaction: a value that starts with a call (dotted names allowed, e.g. pydantic.SecretStr("..."),
    field(default="..."), unwrap_or(x, "...")) has every quoted literal in its argument list
    redacted, not only one directly after the first (; union type annotations
    (str | None = "...") keep their literal defaults redacted (review of #25).
  • CLI: a --dir with no value (trailing, followed by another flag, or empty) is rejected instead
    of falling back to the default workspace; every icx hook fail-open path replies with the
    agent's own no-op for that event (agy: allow on PreToolUse, stop on Stop).
  • Muse adapter review (#27): only the Bash tool's JSON result is unwrapped (other tools' JSON text
    is stored verbatim); subagent identity falls back to child_session_id; a PostCompact
    compact_summary is stored.
  • Deployment guide (review of #26): the ai-memory MCP JSON example carries --server-url and an
    expanded path; the agy wrapper requirement sits next to the global hooks.json example; the
    daemon save-failure text matches v1.2.2.

Known limitations

  • Redaction of quoted secrets inside code-shaped values (chained calls, operator chains, template
    and prefixed literals, lambdas) is incomplete, as in v1.2.2; the redesign is tracked in #29.

Compatibility

  • AgentKind gains value 8 (muse). A v1.2.x binary cannot read a workspace in which a Muse event
    was recorded; upgrade every agent before wiring Muse.

Downloads

Linux x86-64 binaries in two C-library flavours:

  • *-linux-musl: fully static, runs on any Linux distribution.
  • *-linux-gnu: dynamically linked against glibc 2.35 or newer (Ubuntu 22.04,
    Debian 12, RHEL 10 and later); faster in-process hooks than musl.

The x86_64_v3 builds need AVX2 (Intel Haswell / AMD Zen or newer) and exit with
"illegal instruction" on older CPUs; use the baseline x86_64 builds there.
Verify with sha256sum -c SHA256SUMS --ignore-missing or
gh attestation verify <archive>.tar.gz --repo doublegate/Infinity-Context.
Third-party license notices: THIRD-PARTY-NOTICES.txt (also inside each archive).
Corresponding source for every binary is this tag.

icx v1.2.2

Choose a tag to compare

@github-actions github-actions released this 21 Nov 21:28
dafc9b0

Supersedes v1.2.1, which was tagged but never published: a redactor panic was found after
tagging. v1.2.2 contains every v1.2.1 change below plus these fixes (issue #24).

Fixed

  • Redactor panic (denial of service). A quoted literal inside a call that ended in a backslash
    at the end of the input (api_key = Some("abc\) stepped past the last byte and panicked,
    failing that hook and stopping the workspace daemon until the next hook restarted it. The escape
    skip is now bounded to the line, and an exhaustive test runs every combination of up to five
    risky fragments after four sensitive prefixes without a panic.
  • Literals inside nested call wrappers (Some(String::from("...")), Box::new(Secret::new('...')))
    are redacted.
  • Python Optional[str], list[str] and Rust &'a mut T annotations are recognised (they were
    over-redacted); a literal default after one (password: Optional[str] = "...") is still
    redacted, while = None / = True are not secrets.
  • Every subcommand rejects a repeated --dir instead of silently using the last one; icx hook
    still fails open (prints {}).
  • ICXHDC04 tokens are now shape-validated like ICXHDC05 ones before their postings are dropped.

Downloads

Linux x86-64 binaries in two C-library flavours:

  • *-linux-musl: fully static, runs on any Linux distribution.
  • *-linux-gnu: dynamically linked against glibc 2.35 or newer (Ubuntu 22.04,
    Debian 12, RHEL 10 and later); faster in-process hooks than musl.

The x86_64_v3 builds need AVX2 (Intel Haswell / AMD Zen or newer) and exit with
"illegal instruction" on older CPUs; use the baseline x86_64 builds there.
Verify with sha256sum -c SHA256SUMS --ignore-missing or
gh attestation verify <archive>.tar.gz --repo doublegate/Infinity-Context.
Third-party license notices: THIRD-PARTY-NOTICES.txt (also inside each archive).
Corresponding source for every binary is this tag.

icx v1.2.0

Choose a tag to compare

@github-actions github-actions released this 21 Nov 19:16
0f0a878

Remediation and bounded-context release candidate for the independent audit in docs/codex/.
Qualification results and remaining release decisions are tracked in
docs/releases/v1.2.0/.

Migration

  • HDC derived snapshots now use ICXHDC04 with exact token postings. Version 03 remains readable
    and the MMU rebuilds it from preserved CAS events. Older binaries cannot read 04: retain a
    pre-upgrade workspace backup for downgrade. CAS identities and unversioned Q15 fc words do
    not change; incorrect Q1.14 wire labels are superseded by ADR 0006.
  • icx-managed context now defaults to a bounded 32,768-unit policy, reserving 4,096 units each for
    external input/tool schemas and completion. icx context-budget inspects/sets the persisted
    policy. The estimator is a conservative one-UTF8-byte allowance, not a provider tokenizer;
    callers must account for their actual external prompt. Model changes can invalidate prefixes.
  • MCP requires valid JSON-RPC envelopes, typed arguments, an explicit operation and the canonical
    advertised tool name icx. Invalid aliases/arguments now return errors rather than status or
    accidental mutations. Native-hook alias recognition is unchanged.

Fixed

  • YAML sensitive scalars are redacted across literal/folded blocks and indented plain-value
    continuations, including =-containing values, tagged/anchored headers and list mappings.
    Public fields after dedent remain intact. Redaction uses normalized continuation classification
    so the audit's vendor-token/idempotence counterexamples stabilize on the first pass
    (ICX-CODEX-001, ICX-CODEX-006). Existing stored artifacts are not rewritten.
  • Public Profile A skeleton rendering neutralizes raw packet/header delimiters as ASCII bracket
    pairs; standalone parsing rejects unescaped nested markers (ICX-CODEX-002). Existing ordinary
    skeletons and percent literals retain their interpretation. Exact CAS content is unchanged.
  • Exact token postings prevent approximate candidate rejection of stored unique identifiers,
    including multilingual identifiers. Lexical match coverage ranks ahead of HDC/fractal ties;
    common-token queries can require exhaustive refinement (ICX-CODEX-004).
  • Snapshot loads reject nonfinite/out-of-range numerical state and inconsistent cached summaries,
    bundler readouts and 04 posting/Bloom metadata (ICX-CODEX-003).
  • Bus readers reject coherent slots whose sequence differs from the expected ring position;
    extreme public chunker sizes no longer overflow arithmetic (ICX-CODEX-032/033).
  • Daemon success replies follow persistence under the workspace lock; save failure returns an
    error and nonzero daemon result. SQLite WAL uses FULL commit synchronization (ICX-CODEX-030).
  • Codex PostToolUse captures eligible output without emitting the unsupported replacement field.
    Supported prompt/start recall remains; replacement capability is explicitly false (035).
  • Window assembly selects fitting whole recent frames without deleting cold history. Structured
    recall preserves useful handles and leaves unsent sibling-bus events queued (101).
  • Release packaging requires successful trusted CI and REUSE evidence for the exact tagged
    commit, and rechecks before external writes. Missing, partial, failed or expired evidence
    blocks release; a partial failed-job rerun may require a complete workflow rerun (070).
  • Benchmarks measure 1k and 10k scans independently and preserve strict timing-gate failures;
    numerical, corruption, protocol and mutation regressions strengthen assertion coverage
    (072/073). Current specifications supersede historical capability guarantees, and the worked
    U+28C7 example now displays the matching Braille glyph (005/100/102).

Release-review fixes (independent review of the candidate before submission):

  • Redaction: a line ending in a sensitive YAML block header (password: |) no longer exempts
    earlier assignments on the same line ({api_key: x, password: | leaked x). Multi-line plain
    scalars whose continuation lines are less indented than the first, multi-line single- and
    double-quoted scalars, and flow-sequence values (password: [x]) are now hidden. Plural
    collection keys (api_keys, tokens, DB_PASSWORDS) are sensitive; max_passes, bypass
    and addresses are not.
  • Context window: once the budget is reached, history is evicted in large steps (down to half
    the history allowance) behind one stable ICX:HISTORY_OMITTED marker frame, instead of one
    epoch per turn. Consecutive windows now share their prefix between eviction steps (0.94 byte
    reuse at saturation in the 8,192-unit test, previously 0.01); omitted epochs stay
    dereferenceable through the handles in the marker. The eviction point persists with the
    controller state.
  • Lexical index: tokens longer than 256 bytes are not posted, so a long hex/minified run can no
    longer make the persisted ICXHDC04 snapshot undecodable; stored subjects are capped at 64 KiB.
    Ranking weights matched query tokens by inverse document frequency, so a rare exact
    identifier outranks stop-word matches, and very common tokens no longer pull the whole corpus
    into exact scoring (the candidate set is bounded). Snapshot format and version are unchanged.
  • Bus: poll_since_with_loss derives loss from the same head observation as the delivered
    events, so concurrent publishers can no longer make it under-report overwritten events.
  • Redaction no longer hides public digests. The v1.1.0 rule is unchanged (a high-entropy hex
    run of 32+ characters on a line containing = or : is redacted) except for 40-, 64- and
    128-character digests in recognised contexts: sha1:/sha256:/sha384:/sha512:/blake3:
    labels, sha256sum lines, commit <hex> lines, CSV fields and Markdown table cells. Stored
    output can therefore still be checked against SHA256SUMS. Sensitive key names are
    unaffected.
  • YAML/JSON comment and key-separator detection ignores # and : inside quoted values, so a
    colon inside an open quoted secret no longer leaves its continuation lines unredacted and a
    # inside a closed quoted value no longer redacts the following public lines; single-quoted
    values honour the '' escape.
  • A ] inside a quoted element of a sensitive flow sequence (password: ["a]b", "c"]) no
    longer ends redaction early.
  • Text after the closing quote of a multi-line quoted secret is classified like any other line,
    so a key block or a new block/quoted secret opened there is still hidden.
  • The CSV-cell digest exemption no longer applies to quoted elements or flow sequences
    (keys: ["a", "<hex>", "b"] keeps the strict hex rule).
  • icx context-budget --help / -h print usage instead of a missing-value error.
  • The ICX:HISTORY_OMITTED frame uses the same ⟪…⟫ delimiters as the other icx markers.
  • Tests write to a per-user directory under the system temp dir instead of the fixed
    /tmp/Infinity-Context/v120, which could collide between users.
  • Daemon: a workspace-lock timeout or reload failure now fails only that request (the client
    falls back to in-process handling) instead of stopping the daemon. The request is read and
    parsed before the lock is taken, under one 10 s deadline for the whole read, so a slow or
    silent client can no longer hold the workspace lock and starve hooks and mcp-serve.
  • Codex hook replies omit null members ("reason": null, "systemMessage": null); Codex CLI
    0.171 marked hooks returning them as failed.
  • The CLI integration test now runs icx status/deref against a private temporary state
    directory instead of the invoking user's default workspace.
  • Exact parallel HDC scan: icx_hdc::hv::hamming_distances, nearest and par_map_ordered
    split inputs of at least 2,048 vectors across up to 8 scoped threads (no unsafe; results
    identical to a sequential pass, ties to the lowest index). Recall scores large candidate sets
    with it. The icx benchmark 10k gate now measures this scan against a measured 2,000 µs
    ceiling (slowest of 10 quiet-host runs, 1,354 µs, times 1.5); the single-threaded 10k figure
    is still reported, ungated. The previous 1,500 µs ceiling was an unmeasured extrapolation of
    the cache-resident 1k gate, and the 12.8 MB single-threaded scan is memory-bandwidth bound.

Added

  • icx benchmark --json emits measured component values, build flags, target, CPU, fixture hash
    and gate outcomes. scripts/benchmark_series.py records independent processes, warm-up
    exclusion and dispersion; measurements are not end-to-end hook/model guarantees.

  • Release workflow (.github/workflows/release.yml): a version-tag push builds Linux x86-64
    binaries (baseline and x86-64-v3) as static musl and as glibc 2.35+ builds, writes
    SHA256SUMS, attests build provenance and attaches them to a draft GitHub release whose notes
    are the tag's CHANGELOG section. The tagged commit must be on main.

  • Release archives include THIRD-PARTY-NOTICES.txt (generated by cargo-about) with the
    license texts of the statically linked crates, plus musl's notice in the musl builds; the file
    is also attached to each release on its own.

Removed

  • The Unicode and ISO/IEC JTC 1/SC 2/WG 2 publications in
    ref-docs/unicode-braille-patterns/ (the U+2800 code chart, the Unicode 5.2.0 chapter 15
    extract, WG2 N1203/N1279/N1303/N1353/N1453/N1603/N1703/N1770 and UTC 1996-002/027.2) are no
    longer redistributed: their Unicode-3.0 annotation covers Unicode data files, not these
    documents (ICX-CODEX-071). SOURCES.md there lists each document with its official source
    (byte-identical copies verified for three) and the SHA-256 of the copy previously
    distributed; the Unicode Character Database files remain. Git history is unchanged.
  • Raw audit/qualification logs and mutation inventories (42 files, 42.4 MB) moved out of the
    repository into the icx-v1.2.0-raw-evidence.tar.gz release asset, indexed ...
Read more

icx v1.1.0 — audit remediation

Choose a tag to compare

@doublegate doublegate released this 20 Nov 14:27
9d59826

icx v1.1.0 is a remediation release for the internal adversarial audit of v1.0.0 (findings ICX-AUD-001 to ICX-AUD-055, in docs/claude/). Of 51 findings, 42 are fixed, 7 partially fixed, 1 open and 1 deferred by owner decision.

Read the Migration section before upgrading. Three changes are visible to existing users: workspaces move to a private per-project directory under $XDG_STATE_HOME/icx/workspaces/ (a repository-local .icx/ is no longer loaded and must be adopted with icx trust), icx proxy requires a per-install token, and icx deref frames its output as untrusted data (--raw for exact bytes).

Install

Prebuilt Linux x86_64 binaries, built from this tag with cargo build --release --locked on Rust 1.100.0, in two C-library flavours:

Asset C library CPU Notes
icx-v1.1.0-x86_64_v3-unknown-linux-gnu.tar.gz glibc 2.34+ (Ubuntu 22.04, Debian 12, RHEL 9 or newer) x86-64-v3 (AVX2: Intel Haswell / AMD Zen or newer) Recommended on mainstream distributions
icx-v1.1.0-x86_64-unknown-linux-gnu.tar.gz glibc 2.34+ any x86-64 Portable CPU baseline
icx-v1.1.0-x86_64_v3-unknown-linux-musl.tar.gz none (fully static) x86-64-v3 Any distribution, including Alpine and older glibc; passes all 9 icx benchmark gates
icx-v1.1.0-x86_64-unknown-linux-musl.tar.gz none (fully static) any x86-64 Most portable; HDC bind measures ~170 ns/op, above the 100 ns benchmark gate, with no visible effect on hook latency

The glibc archives were added after release by the release workflow (CI run 44843132910) and carry GitHub build-provenance attestations (gh attestation verify <archive>.tar.gz --repo doublegate/Infinity-Context); the musl archives were built locally at release time and are unchanged. SHA256SUMS covers all four archives and THIRD-PARTY-NOTICES.txt, which holds the license texts of the statically linked Rust crates. The glibc archives include that file; for the musl archives, which predate it, it is the separate download.

sha256sum -c SHA256SUMS --ignore-missing
tar xzf icx-v1.1.0-x86_64_v3-unknown-linux-gnu.tar.gz
install -m 0755 icx-v1.1.0-x86_64_v3-unknown-linux-gnu/icx ~/.local/bin/icx
icx init && icx doctor

The v3 builds exit with "illegal instruction" on CPUs without AVX2; use a baseline build there.

Performance of these binaries. The figures elsewhere in these notes and in docs/numbers.md were measured on glibc builds tuned to the measuring machine's CPU. On a 2,000-artifact workspace with a heavier hook workload than the published one (so compare rows, not absolute values), median PostToolUse latency was:

Build In-process (ICX_NO_DAEMON=1) Via icx daemon
glibc, CPU-native (reference) 112.7 ms 10.0 ms
glibc baseline CPU (local build, same configuration as the -gnu baseline archive) 140.4 ms 28.7 ms
musl x86-64-v3 (this release) 142.0 ms 10.7 ms
musl baseline (this release) 169.8 ms 29.5 ms

musl's allocator adds about 20-25% to an in-process hook, and the baseline CPU target roughly triples daemon latency; the glibc v3 archive was not separately measured. Through the daemon the musl v3 binary is within about 7% of the native reference. Run the daemon (ICX_DAEMON_AUTOSTART=1) with these binaries. For the best in-process latency, build from source on the target machine. Corresponding source for every binary is this tag (the source archives below).

From source (any platform with Rust 1.100.0):

git clone --branch v1.1.0 https://github.com/doublegate/Infinity-Context.git
cd Infinity-Context
cargo install --path crates/icx-cli   # Rust 1.100.0 (pinned in rust-toolchain.toml)
icx init && icx doctor

Verification

On the release commit: fmt and clippy (-D warnings) clean, 152 tests passing, line coverage 95.11% (gate 95%), docs build with -D warnings, all 9 benchmark SLO gates pass, cargo-deny, REUSE 3.3 and gitleaks clean. Redactor mutation score 74.7% (68.1% at v1.0.0).

Remediation release for the adversarial audit of v1.0.0 in docs/claude/.
Finding IDs (ICX-AUD-NNN) refer to that audit; per-finding status is in
docs/claude/12-v1.1.0-remediation-status.md.

Migration

  • Workspaces now live under $XDG_STATE_HOME/icx/workspaces/<project-hash>/ in a single
    icx.sqlite3. A repository-local .icx/ is ignored with a warning; adopt one you own with
    icx trust --dir <path>. v1.0.x JSON stores are migrated on first load without deleting them.
  • icx proxy requires the install token: use the base URL it prints
    (http://127.0.0.1:49380/t/<token>) or send x-icx-token. Non-loopback binds need
    --allow-remote.
  • icx deref frames output as untrusted data by default; use --raw for exact bytes.

Security

  • Repository-shipped .icx/ workspaces are no longer loaded into agent context; workspaces are
    owner-checked, 0700/0600, sealed with a per-install key and MAC-verified (ICX-AUD-010, -013).
  • The proxy authenticates clients and binds loopback only by default. The curl header files that
    carry upstream API keys are written to the private runtime directory ($XDG_RUNTIME_DIR/icx)
    and stale ones are swept on start (ICX-AUD-012, -020).
  • icx structural markers anywhere in untrusted input, including the latest message, are escaped
    unless they are byte-identical to a block icx emitted; deref output carries a
    trust=untrusted_data provenance header (ICX-AUD-011).
  • Bounded hook stdin, MCP lines (4 MiB, oversized lines drained without buffering) and fold
    depth (64) (ICX-AUD-014).
  • Only structural tool-call/-result blocks for the icx pager are exempt from marker
    neutralization; tool output that merely contains "name": "icx" is not.
  • Reopening a store resets widened icx.sqlite3, -wal and -shm modes to 0600.
  • install.key is checked (regular file, owner, mode) before it is read, so a FIFO is refused
    instead of blocking; only a missing key is created.
  • icx encode-ahtl refuses an untrusted workspace instead of printing a handle it never stored.
  • icx trust adopts a store's content but not its emitted-block digests, so a planted digest
    cannot exempt forged markers from escaping. MCP grep results carry the same
    trust=untrusted_data provenance header as deref and slice.
  • icx mcp-serve and icx proxy re-verify the workspace per request, so a later icx trust
    takes effect and a workspace whose ownership, mode or seal changed is no longer used.
  • Proxy curl header files go to a directory verified to be ours and private
    ($XDG_RUNTIME_DIR/icx, else <tmp>/icx-<uid>) and stale ones are swept every minute.
  • Redactor: URL userinfo passwords for every scheme, SendGrid, Slack app/webhook, GitLab,
    DigitalOcean, Shopify and Twilio formats, more key names, multi-line values, single-line PEM,
    and bounded unterminated key blocks (ICX-AUD-015, -016). A key with no inline value redacts
    only a following scalar line: nested mappings, comments and prose ending in key: are kept.
    Token-only URL userinfo (https://<token>@host, 16+ mixed or 32+ token characters) is
    redacted. A quoted continuation value
    is treated as a nested key only if its closing quote precedes the separator. Private-key banners are matched
    within one -----BEGIN ...----- banner, and text around a key block is kept.
  • agy transcriptPath opens non-blocking without following symlinks; the bus file refuses
    symlinks, foreign owners and permissive modes (ICX-AUD-018, -019).
  • Hooks never echo input fields back as decisions (Claude Code, Codex and agy PreToolUse; agy
    PostInvocation) and never force an agent to continue (agy Stop) (ICX-AUD-007, -009).

Fixed

  • Hook latency no longer grows with workspace size: SQLite store with append-only rows, persisted
    HDC index snapshot, and a bit-sliced bundler. 338 artifacts: 940 ms to 14 ms per hook in
    process; about 3 ms median through the resident daemon (ICX-AUD-001, -047, -050).
  • Storage is lossless after redaction: CRLF, trailing newlines and literal delimiters survive;
    escaping applies only to model-facing views (ICX-AUD-021).
  • 64-bit handle collisions are refused instead of overwriting, in both the CAS and the DAG,
    including Hermes carried-AHTL lines and subagent summaries (ICX-AUD-022, -023).
  • The daemon flushes and releases the workspace lock at most 1 s after the first unsaved change
    even under continuous hook traffic. A failed flush keeps the unsaved state in memory, releases
    the lock and retries every second; after 5 consecutive failures it logs an error and drops the
    changes so the daemon can exit. Requests that never reach a handler (malformed, wrong
    protocol) do not trigger a save.
  • Concurrent first use of the install key, and concurrent prepare of one workspace, no longer
    collide on temp files or observe a directory before its seal is written.
  • A save refuses to seal a snapshot older than the store on disk (another writer committed
    since it was loaded). If the tables do not hold exactly the in-memory rows (rows removed or
    planted behind the process), they are rewritten from memory before sealing. A new
    workspace's unsaved state survives a first flush that failed after creating the store file. The daemon logs an error when another writer's newer
    store replaces its unsaved changes.
  • Emitted-block digests are stored in an append-only, seal-covered table instead of being
    rewritten inside the state blob on every save.
  • Epoch headers are hex everywhere and epoch ids are u32 (ICX-AUD-024); invalid slice_hint
    values are neither rendered nor parsed (ICX-AUD-025); clause chains are parsed iteratively and
    capped at 64 links (ICX-AUD-026).
  • Associative recall of a unique token in 1,000 log documents: top-1 5.1% to 99.0%, top-5 5.6%
    to 99.7...
Read more