Changelog

All notable changes to aireceipts-cli. Factual, grouped by conventional-commit type (I6: a log, not marketing). Dates are UTC.

v0.8.2 — 2026-07-10

Patch: the auto-attach hook now fires on bare-HEAD branch publishes — git push -u origin HEAD (and --force-with-lease / HEAD:refs/heads/<x> forms), the dominant idiom for coding agents, previously classified as not-a-branch-publish and the SPEC-0073 hook silently wrote no receipt ref. Root-caused from the org dogfood pilots producing zero receipts (issue #228, PR #230): the sibling hook on the same event fired while ours declined classification. The receipt ref slug always derives from the resolved checked-out branch; detached HEAD still attaches nothing (fail-safe), and tag/sha/receipt-ref/delete publishes stay excluded.

Also: docs — terminal-statusline pages now say what happens without tmux (silent fallback + alternatives, #226) and add a shell-wrapper recipe so the statusline appears whenever the agent launches (#225).

v0.8.1 — 2026-07-10

Patch: OpenAI gpt-5.6 family price rows — gpt-5.6-sol ($5.00 in / $0.50 cached / $30.00 out per 1M), gpt-5.6-terra ($2.50 / $0.25 / $15.00), gpt-5.6-luna ($1.00 / $0.10 / $6.00); Standard short-context tier, effective 2026-07-09, every row cited to the official OpenAI pricing page (PR #223). Codex CLI sessions on these models — which rendered tokens-only under I2's no-fabricated-dollars rule — now price in receipts and the statusline. Additive data change only; no code touched.

Also: the README was restructured around a features-first flow with an agent-assisted install section ("point your agent at this README and it installs + wires the statusline"), contributed in PR #213. Docs-only; ships as the npm package README.

v0.8.0 — 2026-07-10

Minor: the statusline now works at the terminal level — tmux, starship, zsh/bash, PowerShell, terminal titles — so Codex and opencode sessions get a live cost line too, not just Claude Code. Neither Codex nor opencode exposes a command-backed status hook (openai/codex#17827 and sst/opencode#8619 are open), so the surface moved to the terminal itself (SPEC-0075, PRs #217/#221).

feat

• aireceipts statusline --cwd <path> selects the newest session attributed to that path instead of the machine's globally newest one, so a tmux status-right line (--cwd "#{pane_current_path}") shows each pane its own session. Attribution is guarded: Claude Code's lossy project-directory encoding is confirmed against the transcript's own recorded cwd after load, sessions launched from the home directory (or /) never ancestor-match everything, dot-segments resolve lexically, full-parse work is capped, and no match renders the neutral placeholder — never another project's line. Cursor sessions carry no cwd and are excluded. Relative paths resolve against the invocation directory; a usable Claude Code stdin payload still wins. Recipes for tmux, starship, plain zsh/bash, PowerShell, and OSC terminal titles are in docs/statusline.md — the Claude Code native statusLine hook remains the recommended (richer, faster) surface where it exists (PRs #217, #221).
• ~/.aireceipts/statusline.json ({"items": ["brand", "cost", …]}) sets a persistent default segment list for the statusline. Precedence: explicit --format > config file > built-in default. An invalid file degrades to the default line with one stderr note (exit 0); the --format flag keeps its fail-fast contract (PR #221).
• Telemetry: --cwd invocations are polling surfaces, so they count locally but never network-flush (a 15-second tmux poll adds zero events). The integration_surface_rendered event gains scoped and configFile booleans — never the path or the format string (docs/telemetry.md, PR #221).

fix

• The published CLI is ~3.5× faster wherever sqlite-backed sessions (opencode, Cursor) are read (PR #214). tsup's removeNodeProtocol default rewrote import("node:sqlite") to import("sqlite") in dist/, which fails at runtime, so every published build silently fell back to spawning the sqlite3 CLI per query — measured 4.0s → 1.15s for one statusline render on a sqlite-heavy machine. On Node ≥ 22.13 the sqlite3 binary is no longer needed at all; older runtimes keep the fallback. Built-artifact regression tests now pin the import specifier and the in-process read path.

chore

• CI gains a windows-latest job (cwd matching, statusline suites, built-CLI e2e). Its first runs surfaced and fixed three real Windows issues: .cmd spawns need a shell under Node's CVE-2024-27980 hardening, goldens/fixtures gained -text line-ending exemptions, and absolute POSIX paths are never drive-prefixed by path.resolve (PR #221).
• OpenSSF Scorecard hardening: esbuild override for the open advisory, absolute SECURITY.md link (PR #212); adopter pr-check workflow hardened with continue-on-error + concurrency groups (PR #207).

Rendered receipt output is byte-identical (receipt goldens unchanged; the help golden gains the --cwd line); the default statusline line renders through the exact same code path as 0.7.2 when neither --cwd nor a config file is present.

v0.7.2 — 2026-07-09

Patch: PR-receipt refs are now written under refs/aireceipts/* instead of the generic refs/receipts/*. The old namespace collided with other tools that also publish to refs/receipts/<slug> (e.g. an attestation producer storing an in-toto receipt.json at the same path). In a repo running both, the namespace was occupied by the other tool's payloads, so CI pr-check fetched a foreign ref, failed its schemaVersion check, and silently posted nothing — the auto-attach hook fired and pushed refs, yet no receipt ever appeared. aireceipts now owns refs/aireceipts/* and never reads or writes refs/receipts/*, so the two coexist; the push classifier still recognizes both namespaces so a foreign receipt-ref push is never mis-counted as a branch commit (PR #204). No migration: aireceipts had never written a refs/receipts/* ref. Hooks and CI on @latest adopt the new namespace together on this release. Rendered receipt output is byte-identical (goldens unchanged); the only user-visible surface change is the ref name in pr --help.

v0.7.1 — 2026-07-09

Patch: Claude Code receipts were over-reporting cost by ~2.5–3× and are now correct. Claude Code writes one assistant transcript record per content block of a response, and every record of that response repeats the same message.id and a byte-identical usage snapshot. The adapter had counted each record as its own billed turn, multiplying a response's tokens by its block count. Turns are now deduped by message.id — the first record of a response books its usage once, later records only add their tool calls — so turnCount and every $ reflect real billed responses (PR #196). After upgrading, every Claude Code receipt you render shows lower dollars and fewer turns than 0.7.0 showed for the same session: a real local session dropped from $102.97 to $36.69 (2.81×), and one PR slice from $5.17 to $1.61. This was a silent over-report — the first known arithmetic case of a receipt exceeding the true floor rather than under-reporting it. Codex, opencode, and Cursor are unaffected (Codex reconciles against the rollout's own cumulative total and was re-verified correct to the cent). Synthetic-fixture output is byte-identical (goldens unchanged); the summary-cache version was bumped (2 → 3) so stale inflated totals recompute on the next run. Receipts posted on this repo's own past PRs carry per-session correction blocks with exact recomputed figures. Full audit and evidence: PR #196, docs/cost-model.md, and the 2026-07-08 addendum in docs/internal/cost-attribution-evidence.md.

v0.7.0 — 2026-07-08

Minor: telemetry now defaults to enabled in CI, reversing the v0.6.x CI-default-off behavior. Automated CI runs (e.g. the pr-check action) are now counted by default, so CI adoption shows up in the data. CI is no longer special-cased — the isCI field still distinguishes CI from human runs. Both kill switches still win everywhere: set AIRECEIPTS_TELEMETRY=off (or 0/false) or DO_NOT_TRACK=1 to disable telemetry in CI or anywhere; an empty/malformed connection string also disables. Telemetry payloads remain content-free (no transcript, prompts, file paths, repo names, or dollar amounts — invariant I4). See SPEC-0002's 2026-07-08 amendment and docs/telemetry.md.

v0.6.1 — 2026-07-08

Patch: the agent auto-attach hook (SPEC-0073) shipped with a 10s timeout — too tight for its first npx -y aireceipts-cli@latest run on a cold npm cache (which downloads the package, including the @resvg/resvg-js native dep, before it can write the ref). On a fresh clone the first push could hit the timeout and silently skip its receipt. Bumped the pre-push hook timeout to 60s (still bounded — the push is never gated on the receipt). The SessionEnd --mini hook timeout is unchanged (10s). No other behavior change; rendered output is byte-identical.

v0.6.0 — 2026-07-08

Minor: receipts can now be produced automatically on push, and PR attribution is more robust. No change to rendered receipt output (goldens byte-identical); both additions are opt-in / internal.

Added

• Agent auto-attach (SPEC-0073): a hidden aireceipts hook pre-push subcommand, wired as a Claude Code PreToolUse hook via a committed .claude/settings.json, writes and pushes the receipt ref (refs/receipts/<slug>) when the coding agent runs a branch push — so a receipt is produced with no manual command. It never blocks the push (exits 0 on every path, no stdout, bounded by the hook timeout) and only acts on a real origin/current-branch push (tokenized detection; skips dry-run/delete/tags/non-origin/repo-retargeting/refs/receipts). The adopter kit is now two files — the pr-check workflow (produces nothing on its own) plus the .claude/settings.json hook (docs/adopt/, aireceipts integrations).

Fixed

• Robust PR-receipt attribution (SPEC-0072): a session whose commit SHA was orphaned by git commit --amend (or rebase) is now correctly credited by matching its stable git patch-id to the branch commit, instead of showing "no branch commit" while an unrelated helper takes the credit. Strictly under-crediting — duplicated diffs, empty/merge commits, and a diff already directly claimed by another session never produce a match (no fabricated or stolen credit, I2). Adds an unanchored-git-write confidence signal that floors the total when a real git-write can't be tied to the branch.

Minor: PR receipts gain a self-contained CI path (no external reusable workflow), the default statusline gets richer, and the samosa tip link leaves the default PR-posted surfaces (now opt-in). Two deliberate output changes — statusline default and the PR-posted footer — both called out below.

Added

• Self-contained pr-check (SPEC-0064 R2–R7): an adopter's own workflow can run npx -y aireceipts-cli@latest pr-check in a single job — no reusable-workflow uses:, no org Actions-policy gate — to fetch the branch's receipt ref, render + sanitize it, and upsert the marked PR comment via GITHUB_TOKEN (no local gh). Hidden command; for trusted same-repo/internal PRs (the two-job reusable workflow stays the hardened path for untrusted fork PRs). Also fixes the reusable-caller template, which was missing pull-requests: write.

Changed (deliberate output changes)

• Richer default statusline (SPEC-0071): the default line now carries a burn rate ($/hr), context-window %, abbreviated M/B token counts, and an inline 5h reset countdown — e.g. [aireceipts] $423 · $80/hr · 501M · ctx 42% · 5h 26% ↺2h13m. Every segment renders from a real payload field or a priced ledger value; nothing is estimated.
• Samosa tip link off by default (SPEC-0070): the buy me a samosa link no longer appears on the surfaces aireceipts posts onto a PR (the comment's <details> and the artifact footer). Opt back in with --samosa; the standalone tip page and the docs-site footer are unchanged.

Internal / CI

• Incremental mutation testing (SPEC-0069): Stryker runs incrementally on PRs that touch the money paths (src/pricing/**, src/pr/**), with a nightly full sweep — the same anti-gaming moat at a fraction of the wall-clock. No product code change.

v0.4.0 — 2026-07-07

Minor: PR receipts can now attach and post themselves. A receipt travels with the branch as a sibling git ref and CI renders + posts the comment, so a receipt lands even when no one runs pr --post. Rendered receipt output is unchanged from v0.3.0 (goldens byte-identical); the new surface is opt-in.

Added

• Seamless PR receipts — store=ref + pre-push hook (SPEC-0065): aireceipts pr --store ref (or AIRECEIPTS_STORE=ref) writes the receipt as a deterministic git object on refs/receipts/<slug> — invisible in the tree and PR diff, byte-identical across machines (dated from the session's own endedAt, never wall-clock). A committed .githooks/pre-push hook generates and pushes that ref on git push (best-effort; never blocks a push, no second push). npm run setup:hooks activates the hook for a clone of this repo; the default store stays comment.
• CI posts the receipt from the ref (SPEC-0066): a two-job pr-receipt-check.yml fetches the branch ref, validates and sanitizes the untrusted, fork-author-controlled payload, renders it through the existing renderer, and upserts one marker comment via GITHUB_TOKEN — no local gh required. Opt-in enforcement (AIRECEIPTS_REQUIRE_PR_RECEIPT=true) makes same-repo PRs require a receipt; fork PRs always stay notice-only (their transcripts live on the contributor's machine, so CI can't generate one). Enforcement is currently coarse — it does not yet distinguish agent-built from hand-written PRs.
• @latest distribution pin (SPEC-0064 R1): the PR-receipt-check caller installs the reusable workflow at @latest; the moving latest git tag is advanced by the release workflow on each publish. (SPEC-0064 R2–R4, the npm-native pr-check command, remain in progress.)

Performance

• Faster feedback (SPEC-0063): in-process sqlite parsing, a goldens compile cache, and parallel preflight reduce CLI and preflight wall-clock.

Security

• The receipt payload that rides on the branch ref is treated as untrusted input — anyone who can push a branch, including a fork author, controls its bytes. The CI post path is a hardened trust boundary: a field-aware sanitizer (bracket-escape + autolink defang for live-markdown fields, fence-guard for fenced text, an https-allowlist that rejects markdown-breaking characters for artifact URLs), rendered in a token-less render job whose data-only artifact is posted by a separate post job — golden-gated against an injection corpus.

Docs

• Adopter kit: a minimally-invasive default plus opt-in tiers for repos that want to turn enforcement up (docs/pr-receipts.md, docs/adopt/).

v0.3.0 — 2026-07-06

Minor: background-agent (subagent) spend becomes visible on every surface, the statusline is rebranded and extensible, and receipts get sharper defaults plus retroactive/first-run tooling. Per the 0.x policy, this minor deliberately changes rendered output — the callouts below list every change.

Output changes (deliberate, golden-gated)

• Statusline prefix is now [aireceipts] in stdin mode and [aireceipts · <agent>] in disk-fallback mode (was [Claude Code] etc.), and the default line appends your official 5h quota window (5h N%) when Claude Code's payload carries it (SPEC-0062).
• Session receipts draw one SUBAGENTS (N) row (with TOTAL covering it) when the session has child transcripts; SPEC-0061 itself adds nothing to childless sessions (their v0.2.0→v0.3.0 rendering differences come only from the SPEC-0054/0055 changes above).
• The receipt footer now carries the install CTA (npx aireceipts-cli) instead of the samosa link, and the inline methodology paragraph moved behind --methodology (SPEC-0055).
• The same tokens on <model> line gains a (N% less) suffix when it is real savings (SPEC-0054).

Added

• Subagent rollups everywhere (SPEC-0060, SPEC-0061): PR comments aggregate each contributor's subagents into one fence row plus a capped details table; session receipts, the statusline, the install-hook mini-receipt, and --json (optional subagents object) fold the same priced atoms in, with floor caveats for anything unreadable or unpriced — never a fabricated $.
• Statusline v2 (SPEC-0062): --format <segments> engine (brand,cost,tokens,waste,quota5h,quota7d,quotaEta); quotaEta is a labeled ≈ cap-crossing estimate from two observed readings, rendered only when its guards hold (state file: ~/.aireceipts/quota-window.json).
• --details (SPEC-0054): opt-in receipt section — token/cache anatomy, turns, peak turn, cache-read repricing, BY MODEL split.
• backfill (SPEC-0056): bulk retroactive receipts for every existing session on disk.
• --demo (SPEC-0051): a bundled sample session renders a real receipt on a machine with no sessions yet; both empty-state messages point at it.
• setup + integrations (SPEC-0050): first-run report and exact local integration snippets for Claude Code, Codex, opencode, Cursor, and GitHub.
• Savings slip (SPEC-0059): could-have-saved handoff block and PR section.
• --version, NOTICE file, and the OIDC release pipeline (#134); OpenSSF Scorecard, community files, and CI telemetry default-off (#139).
• Per-agent docs pages for the five supported agents (SPEC-0058); discovery now flags unreadable sessions instead of silently skipping them (SPEC-0045).

Fixed

• Weekly-digest delta-direction labels now match the sign of the change (SPEC-0008, #150).
• spec-lint catches duplicate spec ids across filenames (#136).

Docs

• README rebuilt PR-receipt-first with a real merged-PR receipt comment as the hero, then trimmed to one receipt showing per format (SPEC-0053, #153).
• Positioning pass across README/FAQ/landing (SPEC-0048/0049); Show HN launch kit and GTM sequencing docs; statusline docs cover the new segments and the host refresh-cadence limitation.

v0.2.0 — 2026-07-05

Minor: adds adoption telemetry + a local stats command (SPEC-0043), and lands the security/privacy hardening from the v0.1.0 release-board findings (which were live in v0.1.1).

Added

• Adoption telemetry v2 (SPEC-0043): a nine-event catalog (feature-usage events, activation milestones), a pseudonymous random-UUID install identity sent only as a salted hash, and a local receipts counter — all content-free, opt-out, and disclosed in the updated first-run notice. (#110)
• aireceipts stats — a new command that prints your local receipts-generated / total-runs / first-run counts from ~/.aireceipts/state.json; works fully offline and even with telemetry disabled, and never leaves your machine. (#110)

Fixed

• PR-receipt cost confidence — the implemented slices of SPEC-0044 (the spec stays building: its --self-check kill-criterion and cost-model docs are still pending, so it is not flipped to shipped):
• Every contributor drop/degrade/lower-bound now routes through a typed ConfidenceEvent and surfaces in the PR receipt — no more silent drops (the mirror of the #87 over-credit bug); a compile-time exhaustive switch + a hygiene guard make "no silent wrongness" a property. (#117)
• Oracle-independent cost matrix (scenario × agent). (#120)
• Receipt rows now sum to the displayed total. (#121)
• Cache-write lower-bound caveat, only when actually under-priced. (#122)
• Silent parse-skip and load-failure drops surfaced. (#123)
• Grandchild subagent counted once, not twice. (#124)
• src/pr/** mutation-gated + the silent-drop guard broadened. (#126)
• Strip terminal escape sequences (ANSI/CSI/OSC/nF) and C0/C1 control characters from all transcript-derived display text — session titles, tool names, and model-mix labels — across every adapter. A crafted transcript could previously emit raw escapes to the operator's terminal (e.g. recolor output or retitle the window via OSC-0). (#112)
• aireceipts --list --json on zero sessions now emits valid JSON [] on stdout with the message on stderr, instead of plain text that broke | jq. (#112)
• aireceipts --telemetry-show — the command that previews what telemetry would be sent — no longer records or flushes a cli_run event itself; previewing telemetry sent telemetry. (#115)
• Bump the summary-cache version so titles cached by the pre-sanitizer parser are re-parsed rather than served raw. (#112)

Docs

• Correct README/getting-started privacy and coverage claims (transcripts/code never uploaded; diagnostics are opt-out; supported-agent list is finite); sync docs/telemetry.md agent-type enums and kill-switch examples; correct the week --json and source entries in docs/json-schema.md; drop a stale pre-release note. (#115)

Chore

• Restructure the opencode combinatorial unit test: validate all summaries from one listSessions() and deep round-trip only a structural-coverage sample, instead of reopening the SQLite DB per session (6m40s → ~20s, coverage retained). Local preflight-release.mjs sets AIRECEIPTS_SKIP_STRESS=1 so the spawn-heavy 100-session e2e stress case doesn't wedge on throttled dev macOS; CI runs the full suite (env unset), so coverage is unchanged.
• Ledger: SPEC-0040/0041/0042 flipped to shipped (they shipped in v0.1.1); AGENTS.md current-state inventory brought up to date; CHANGELOG.md added.

v0.1.1 — 2026-07-04

First release through the OIDC trusted-publishing workflow (v0.1.0 was the manual bootstrap).

Added

• Parse Codex compaction records (compacted + context_compacted) into the normalized model, so context-thrash can fire on Codex sessions. (SPEC-0040)
• Real-session discovery filter: exclude workflow-journal artifacts under subagents/ from listings; floor all-zero artifacts out of aggregate windows. (SPEC-0041)
• --handoff resume packet: a deterministic state header, a covers: line, and a versioned --handoff --json surface. (SPEC-0042)

Chore

• Size two long-running SQLite test timeouts for macOS background-QoS throttling of vitest-spawned children; CI unaffected. (#111)

v0.1.0 — 2026-07-04

Initial public release: the receipt engine and its full surface (parse adapters, cited price tables, per-tool attribution, waste lines, compare, week, budget, handoff, PR receipts, SVG/PNG export, templates, disclosed opt-out telemetry).