CLI: surrogate-safe tail truncation for task-verify failure output and failure receipts #860

Self-discovery (undirected maintainer review pass). Left unclaimed for the next natural tick.

sourceType

Self-discovery (repository maintainer review of qwen-code-dev-bot/oh-my-cli).

sourceLinkOrEvidence

• src/task-verify.ts:74-76 — scrubOutput guards with scrubbed.length <= MAX_OUTPUT_BYTES (UTF-16 code units vs a constant named BYTES) then keeps the tail via scrubbed.slice(scrubbed.length - MAX_OUTPUT_BYTES).
• src/task-verify.ts:83 — displayCommand head-truncates with a raw text.slice(0, MAX_COMMAND_LEN).
• src/failure-receipts.ts:98-100 — tail() returns "…" + text.slice(text.length - maxChars).
• src/text-cut.ts — every helper is a head cut (safeCutEnd L35, dropLastCodePoint L49, safeByteCutEnd L65, clampMarked L78). There is no surrogate-safe tail-cut helper.

Minimal reproduction (run with node):

function tail(text, maxChars){ return text.length <= maxChars ? text : "…" + text.slice(text.length - maxChars); }
tail("aaaa\u{1F680}bbbb", 5);            // -> "…" + slice(5) starts at the LOW surrogate (0xde80) => orphaned
const MAX = 8*1024, s = "\u{1F680}".repeat(3000);
s.length <= MAX ? s : s.slice(s.length-MAX); // 6000 code units pass the guard but are 12000 UTF-8 bytes (> 8192)
("a".repeat(MAX-1)+"\u{1F680}"+"a".repeat(MAX-1)).slice(("a".repeat(MAX-1)+"\u{1F680}"+"a".repeat(MAX-1)).length - MAX); // tail begins 0xde80

Observed: lone low surrogate 0xde80 at the start of the kept tail, and the 8 KiB budget exceeded by ~47%.

problemStatement

Two display paths keep the tail of untrusted text via a raw String.prototype.slice(length - N). When the cut lands inside a surrogate pair, the kept tail begins with an orphaned low surrogate, producing malformed output. task-verify additionally compares/slices by UTF-16 code units against a constant named MAX_OUTPUT_BYTES, so multibyte output can exceed the intended 8 KiB byte budget. The head-cut direction was hardened long ago (#856 and the #812–#858 series), but text-cut.ts has no tail-cut helper, so these tail sites were never covered.

userValue

Failure output shown when a --verify command fails, and persisted failure receipts, render as well-formed text instead of text containing orphaned surrogates; task-verify's captured output actually honors its 8 KiB bound for multibyte content.

scope

• Add a surrogate-safe tail-cut helper to src/text-cut.ts (returns a start index such that text.slice(start) never begins with an orphaned low surrogate), plus a byte-budget variant for task-verify.
• src/task-verify.ts:scrubOutput — keep the tail within MAX_OUTPUT_BYTES bytes without splitting a surrogate.
• src/task-verify.ts:displayCommand — use safeCutEnd for the MAX_COMMAND_LEN head cut.
• src/failure-receipts.ts:tail — use the surrogate-safe tail cut so the …-prefixed tail never starts with an unpaired surrogate.
• Red→green regression tests for each site and the new helper.

nonGoals

• No change to existing head-cut helpers' behavior (safeCutEnd, safeByteCutEnd, clampMarked).
• No change to failure-receipt retention limits or FAILURE_OUTPUT_TAIL_CHARS values.
• Not re-litigating the already-closed head-truncation sweep (Robustness: make ellipsis truncation surrogate-safe — never split an emoji/astral character at the cut boundary #812/Robustness: complete surrogate-safe ellipsis truncation across the ~26 remaining raw-string sites (follow-up to #812) #821/Robustness: make the remaining ellipsis truncations surrogate-safe (event detail + fatal message) #826/Robustness: make byte-bounded truncations surrogate-safe (headless output + change-review path/name) #828/Robustness: make the hook output-size cap surrogate-safe (spawnHookRunner) #830/Robustness: make the tool-invocation output-size cap surrogate-safe (spawnCommandRunner) #832/Robustness: account subprocess output caps in UTF-8 bytes, not UTF-16 code units (hook/tool/MCP) #834/Robustness: reassemble multi-byte UTF-8 split across chunks in the hook/tool/MCP output caps #836/Robustness: surrogate-safe truncation for subagent result and side-question content clamps #840/Robustness: surrogate-safe truncation for tui-shell tool-text rendering (transcript, output, hints) #842/Robustness: make the provider-invocation output cap UTF-8-byte-accurate and surrogate-safe (capOutput) #848/Robustness: session-export clamp() splits surrogate pairs — exported transcripts can carry an unpaired surrogate at the truncation boundary #854/Robustness: invocation one-line display collapse() splits surrogate pairs — provider/tool/MCP Result/Stdout/Stderr can carry an unpaired surrogate #856/Test: add surrogate regression tests for the provider/MCP display-collapse paths (gap left by #856) #858).
• Other raw-slice sites outside the tail/head paths named above (e.g. workspace-memory, session-goal) are out of scope here.

acceptanceCriteria

• Given combined --verify output whose tail would straddle a surrogate pair at the 8 KiB boundary, scrubOutput returns text with no orphaned surrogate (regex [\ud800-\udbff](?![\udc00-\udfff])|(?<![\ud800-\udbff])[\udc00-\udfff] does not match) and Buffer.byteLength(result) <= MAX_OUTPUT_BYTES.
• Given a command longer than MAX_COMMAND_LEN containing an astral character at the boundary, displayCommand produces no orphaned surrogate.
• Given failure-receipt stdout/stderr whose tail straddles a surrogate pair at FAILURE_OUTPUT_TAIL_CHARS, tail() returns text with no orphaned surrogate (the … prefix is preserved).
• The new text-cut.ts helper is exported and unit-tested; existing head-cut behavior is unchanged.

testPlan

• Unit tests for the new tail helper: tail boundary straddling a pair, tail fully in-range, long-ASCII tail, byte-budget multibyte case.
• Regression tests in tests/unit/task-verify.test.ts (multibyte scrubOutput, boundary displayCommand) and tests/unit/failure-receipts.test.ts (surrogate-straddle tail).
• Red→green proof: reverting the fixes to raw slices must make the new tests fail; restoring makes them pass.
• Gates: npm run typecheck, npm run build, npm test (unit), npx vitest run tests/integration, npm run smoke, npm audit --omit=dev --audit-level=high.

dogfoodPlan

After merge, run npx vitest run tests/unit/task-verify.test.ts tests/unit/failure-receipts.test.ts tests/unit/text-cut.test.ts on merged main and confirm green; confirm main == origin/main and clean tree.

riskAndSecurityNotes

Tests-only behavior surface otherwise unchanged. The inputs are untrusted command output, so fixes must preserve redaction (redactSecrets) and path-scrubbing order; tests must not weaken those. No new secrets/credentials introduced.

duplicateSearchEvidence

• gh search issues "surrogate" → only the closed head-cut series Robustness: make ellipsis truncation surrogate-safe — never split an emoji/astral character at the cut boundary #812/Robustness: complete surrogate-safe ellipsis truncation across the ~26 remaining raw-string sites (follow-up to #812) #821/Robustness: make the remaining ellipsis truncations surrogate-safe (event detail + fatal message) #826/Robustness: make byte-bounded truncations surrogate-safe (headless output + change-review path/name) #828/Robustness: make the hook output-size cap surrogate-safe (spawnHookRunner) #830/Robustness: make the tool-invocation output-size cap surrogate-safe (spawnCommandRunner) #832/Robustness: account subprocess output caps in UTF-8 bytes, not UTF-16 code units (hook/tool/MCP) #834/Robustness: reassemble multi-byte UTF-8 split across chunks in the hook/tool/MCP output caps #836/Robustness: surrogate-safe truncation for subagent result and side-question content clamps #840/Robustness: surrogate-safe truncation for tui-shell tool-text rendering (transcript, output, hints) #842/Robustness: make the provider-invocation output cap UTF-8-byte-accurate and surrogate-safe (capOutput) #848/Robustness: session-export clamp() splits surrogate pairs — exported transcripts can carry an unpaired surrogate at the truncation boundary #854/Robustness: invocation one-line display collapse() splits surrogate pairs — provider/tool/MCP Result/Stdout/Stderr can carry an unpaired surrogate #856/Test: add surrogate regression tests for the provider/MCP display-collapse paths (gap left by #856) #858; none covers tail truncation, task-verify, or failure-receipts.
• gh search issues "task-verify" → only Task verification: run the canonical verify commands and emit a bounded, redacted, head-bound pass/fail verdict #66 (original feature, closed); no truncation issue.
• gh search issues "receipt" → only CLI: bounded redacted failure receipts for shell tool runs, viewable per session #574 (original failure-receipts feature, closed); no truncation issue.
• Verified src/text-cut.ts contains no tail-cut helper today.

parentChildRelationship

Standalone leaf subtask; no parent roadmap item required and no children anticipated.

dependencyOrder

No dependencies; dependency-ready. Self-contained change to text-cut.ts, task-verify.ts, failure-receipts.ts + tests.