Rectify: Stdout Idle Watchdog + Bounded Suppression for Stall Detection by Trecek · Pull Request #735 · TalonT-Org/AutoSkillit

Trecek · 2026-04-11T22:04:23Z

Summary

A headless Claude subprocess hung for 82 minutes on a silently stalled HTTPS SSE stream. The TCP connection remained ESTABLISHED with zero traffic, so the _has_active_api_connection suppression gate in _process_monitor.py kept resetting the stale-kill timer every 2-second poll cycle. Neither Channel A nor Channel B detected the stall because Channel A only checks for valid JSONL result records (not raw byte growth), and Channel B's stale detection was suppressed by the ESTABLISHED port-443 connection.

Three layers of fix provide immunity:

Stdout idle watchdog (primary) — new _watch_stdout_idle coroutine in the anyio task group monitors stdout temp file byte count. Fires TerminationReason.IDLE_STALL if no growth for idle_output_timeout seconds. NOT suppressed by active API connections — orthogonal to Channel A/B.
Bounded suppression (defense-in-depth) — max_suppression_seconds caps the _has_active_api_connection and _has_active_child_processes suppression gates so they cannot defer stale kills indefinitely. Counter resets on genuine JSONL activity.
Subprocess timeouts — timeout= added to all unguarded subprocess.run() and asyncio subprocess calls in remote_resolver, smoke_utils, clone, and token_summary_appender.

Changed Files

New (★)

tests/execution/test_process_idle_watchdog.py — watchdog coroutine unit tests
tests/workspace/test_clone_timeouts.py — static analysis: git network commands have timeouts

Modified (●)

src/autoskillit/core/_type_enums.py — IDLE_STALL members on TerminationReason and CliSubtype
src/autoskillit/config/settings.py — idle_output_timeout and max_suppression_seconds on RunSkillConfig
src/autoskillit/execution/_process_race.py — _watch_stdout_idle coroutine, idle_stall field on RaceAccumulator/RaceSignals, priority update in resolve_termination
src/autoskillit/execution/_process_monitor.py — bounded suppression logic with suppression_start tracker
src/autoskillit/execution/process.py — wire watchdog into task group, IDLE_STALL kill dispatch
src/autoskillit/execution/headless.py — IDLE_STALL routing in _build_skill_result (retriable)
src/autoskillit/execution/session.py — exhaustive match arms for IDLE_STALL
src/autoskillit/core/_type_subprocess.py — protocol updated with new params
src/autoskillit/execution/recording.py — runner implementations updated
src/autoskillit/execution/remote_resolver.py — 15s timeout via asyncio.wait_for
src/autoskillit/smoke_utils.py — timeout=60 on all subprocess calls
src/autoskillit/workspace/clone.py — timeout=300 (clone), timeout=120 (push)
src/autoskillit/hooks/token_summary_appender.py — timeout=30 on gh api calls

Test Plan

Implementation Plan

Plan file: .autoskillit/temp/rectify/rectify_idle_stall_watchdog_2026-04-11_140500.md

🤖 Generated with Claude Code via AutoSkillit

Trecek

AutoSkillit PR Review — Verdict: changes_requested (posted as COMMENT — cannot REQUEST_CHANGES on own PR)

Trecek · 2026-04-11T22:30:01Z

-            stdout, _ = await proc.communicate()
+            io_task = asyncio.ensure_future(proc.communicate())
+            try:
+                await asyncio.wait_for(proc.wait(), timeout=15.0)


[warning] defense: wait_for(proc.wait()) races with concurrent proc.communicate() future. wait() does not drain stdout pipe; if subprocess fills the pipe buffer before exiting, this can deadlock. Use asyncio.wait_for(proc.communicate(), timeout=15.0) directly.

Investigated — this is intentional. The code creates io_task = asyncio.ensure_future(proc.communicate()) at line 54 BEFORE calling wait_for(proc.wait()). The communicate() future runs concurrently on the event loop and actively drains stdout, so the pipe buffer deadlock scenario cannot occur here. wait() alone would deadlock, but the concurrent communicate() task prevents it.

Trecek · 2026-04-11T22:30:01Z

        input_data: str | None = None,
        completion_drain_timeout: float = 5.0,
        linux_tracing_config: Any | None = None,
+        idle_output_timeout: float | None = None,


[warning] arch: idle_output_timeout and max_suppression_seconds are execution-tuning params (L1 concerns) added to the HeadlessExecutor protocol in L0 core/. This couples L0 to L1 operational details. Consider kwargs escape hatch or moving protocol to L1.

Investigated — this is intentional. The SubprocessRunner protocol already contained multiple execution-tuning params at L0 before this PR: timeout (float), stale_threshold (float=1200), completion_drain_timeout (float=5.0), pty_mode (bool), linux_tracing_config (Any|None). The new idle_output_timeout and max_suppression_seconds follow the identical pre-existing pattern. This is the established design for this protocol — it serves as the structural contract mirroring run_managed_async.

Trecek · 2026-04-11T22:30:01Z

+    Orthogonal to Channel A/B: NOT suppressed by active API connections.
+    Monitors raw byte count (st_size), not JSONL record structure.
+    """
+    import time as _time


[info] cohesion: import time as _time is a local import inside _watch_stdout_idle. The rest of the module uses module-level imports. Promote to module level.

Trecek · 2026-04-11T22:30:01Z

+        await anyio.sleep(_poll_interval)
+        try:
+            current_size = stdout_path.stat().st_size
+        except OSError:


[info] defense: except OSError: continue silently swallows all OS errors on stdout_path.stat(). ENOENT is expected, but other errors (permission denied, I/O error) should be logged at DEBUG.

Trecek · 2026-04-11T22:30:01Z

-        _phase2_poll=_phase2_poll,
-        _phase1_timeout=_phase1_timeout,
-        expected_session_id=acc.stdout_session_id,
+        **_monitor_kwargs,  # type: ignore[arg-type]


[info] defense: # type: ignore[arg-type] suppresses type error without rationale. dict[str, object] is too wide for the target signature. Consider TypedDict or add explanatory comment.

Trecek · 2026-04-11T22:30:01Z

    completion_drain_timeout: float = 5.0
    exit_after_stop_delay_ms: int = 120000
+    idle_output_timeout: int = 600
+    max_suppression_seconds: int = 1800


[info] defense: max_suppression_seconds and idle_output_timeout have no range validation. Zero or negative values would cause silent logic bugs (immediate expiry). Consider post_init assertion.

Trecek · 2026-04-11T22:30:01Z

+            is_error=True,
+            exit_code=-1,
+            needs_retry=True,
+            retry_reason=RetryReason.STALE,


[info] defense: RetryReason.STALE reused for IDLE_STALL termination. Downstream retry telemetry cannot distinguish the two failure modes.

…ss timeouts Addresses 82-minute hung SSE stream where ESTABLISHED TCP connection suppressed stale-kill indefinitely. Three layers of fix: 1. Stdout idle watchdog (_watch_stdout_idle): independent coroutine in the anyio task group that monitors stdout temp file byte count. Fires IDLE_STALL if no growth for idle_output_timeout seconds. NOT suppressed by active API connections — orthogonal to Channel A/B. 2. Bounded suppression: max_suppression_seconds caps _has_active_api_connection and _has_active_child_processes suppression gates so they cannot defer stale kills indefinitely. Counter resets on genuine JSONL activity. 3. Subprocess timeouts: add timeout= to all unguarded subprocess.run() and asyncio subprocess calls in remote_resolver, smoke_utils, clone, and token_summary_appender.

…IDLE_STALL

Add io_task.cancel() in the TimeoutError handler to prevent a task leak when proc.communicate() future is abandoned after timeout. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Split shared suppression_start into suppression_start_api and suppression_start_child to prevent elapsed time carry-over when sessions alternate between API connection and child process conditions. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

…ynaconf The contract test test_every_subconfig_field_referenced_in_from_dynaconf requires every RunSkillConfig field to have a val() line in from_dynaconf. These two fields were added to the dataclass but not wired into the config builder, causing merge-queue CI failure. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

## Summary `clone_repo` collapses three distinct subprocess outcomes (`TimeoutExpired`, non-zero `returncode`, genuinely absent origin) into a single silent fallback — `detected_url = ""` → `_resolve_clone_source` → `git clone <local-path>` via git's local transport. The return dict's `remote_url` is an untagged `str`, so four downstream recipes that splice `context.remote_url` into shell commands and `push_to_remote` cannot distinguish a healthy clone from a silently-corrupted one. The same `""` value causes the `if effective_url:` guard at `clone.py:333` to skip the `#377` origin-isolation rewrite, compounding the regression. **Bug class:** "subprocess fails → value collapses to empty string → downstream consumers silently use the empty value → invariant violation goes undetected." Five prior PRs (#368, #377, #424, #544, #735) each patched an adjacent sub-path but left this collapse site untouched. **Part A architectural immunity (three phases, sequentially layered):** 1. **Typed probe result.** Replace the one-line `_resolve_clone_source` collapse with `_probe_origin_url` returning a `CloneSourceResolution` dataclass carrying `reason` (`"ok" | "no_origin" | "timeout" | "error"`), `url`, and `stderr`. The call site must switch on `reason`; no path can collapse the four cases back into a single string. 2. **Typed return contract.** Introduce `CloneResult` — a `TypedDict` union with a `clone_source_type: Literal["remote", "local"]` discriminator on the success shape. mypy statically rejects any return branch that omits the discriminator, so provenance is impossible to lose. 3. **Unconditional origin isolation.** Extract `_ensure_origin_isolated` and call it for every successful clone regardless of URL availability. Removes the `if effective_url:` guard that silently skipped the `#377` rewrite. Together these three phases make the `git clone <local-path>` failure mode structurally unreachable from the default strategy: the only way to get a local-source clone is the explicit `strategy="clone_local"` branch (`shutil.copytree`), which now also produces a typed success result with `clone_source_type="local"` and a `file://` origin. ## Architecture Impact ### Error/Resilience Diagram ```mermaid %%{init: {'flowchart': {'nodeSpacing': 40, 'rankSpacing': 55, 'curve': 'basis'}}}%% flowchart TB %% CLASS DEFINITIONS %% classDef terminal fill:#1a237e,stroke:#7986cb,stroke-width:2px,color:#fff; classDef stateNode fill:#004d40,stroke:#4db6ac,stroke-width:2px,color:#fff; classDef handler fill:#e65100,stroke:#ffb74d,stroke-width:2px,color:#fff; classDef phase fill:#6a1b9a,stroke:#ba68c8,stroke-width:2px,color:#fff; classDef newComponent fill:#2e7d32,stroke:#81c784,stroke-width:2px,color:#fff; classDef output fill:#00695c,stroke:#4db6ac,stroke-width:2px,color:#fff; classDef detector fill:#b71c1c,stroke:#ef5350,stroke-width:2px,color:#fff; classDef gap fill:#ff6f00,stroke:#ffa726,stroke-width:2px,color:#000; START([START]) %% ── RECIPE-TIME VALIDATION ───────────────────────────────────────────── subgraph RecipeValidation ["★ RECIPE-TIME VALIDATION (rules_clone.py)"] direction TB RuleGate["★ clone-local-strategy-with- remote-url-capture ━━━━━━━━━━ fired at open_kitchen / load_recipe inspects strategy + capture_list"] RuleErr["★ RuleFinding ERROR ━━━━━━━━━━ strategy='clone_local' + remote_url capture"] RuleWarn["★ RuleFinding WARNING ━━━━━━━━━━ templated / unknown strategy + remote_url capture"] end T_RECIPE_BLOCKED(["RECIPE_BLOCKED recipe invalid at load time"]) %% ── CLONE ENTRY GATES ────────────────────────────────────────────────── subgraph CloneGates ["● CLONE ENTRY GATES (clone_repo)"] direction TB SourceCheck{"● source_dir exists?"} UncommittedGate{"● detect_uncommitted_changes ━━━━━━━━━━ only when strategy='' never raises"} UnpublishedGate{"● detect_unpublished_branch ━━━━━━━━━━ 15 s timeout — fail-open no origin → False"} end T_VAL_ERR(["ValueError source_dir missing"]) T_UNCOMMITTED(["★ CloneGateUncommitted ━━━━━━━━━━ soft stop: dirty tree returns dict, no raise"]) T_UNPUBLISHED(["★ CloneGateUnpublished ━━━━━━━━━━ soft stop: branch not on remote returns dict, no raise"]) %% ── ORIGIN PROBE ─────────────────────────────────────────────────────── subgraph OriginProbe ["★ ORIGIN PROBE (_probe_origin_url)"] direction LR Probe["★ git remote get-url origin ━━━━━━━━━━ 30 s timeout always runs (all strategies)"] ProbeOK["★ reason='ok' URL captured"] ProbeNoOrigin["★ reason='no_origin' no remote configured"] ProbeTimeout["★ reason='timeout' subprocess hung ≥ 30 s"] ProbeError["★ reason='error' non-zero / unexpected stderr"] end %% ── STRATEGY ROUTING ─────────────────────────────────────────────────── subgraph StrategyRouter ["● STRATEGY ROUTING (clone_repo)"] direction TB StratDecide{"● strategy?"} ProbeOKCheck{"probe.reason == 'ok'?"} LocalCopy["clone_local: ━━━━━━━━━━ shutil.copytree remote_url always ''"] RemoteClone["● git clone {url} ━━━━━━━━━━ 300 s timeout --branch if set"] end T_PROBE_ERR(["RuntimeError clone_origin_probe_failed use strategy='clone_local' for local-only repos"]) T_CLONE_FAIL(["RuntimeError git clone non-zero exit"]) %% ── POST-CLONE ISOLATION ─────────────────────────────────────────────── subgraph PostClone ["POST-CLONE ORIGIN ISOLATION (_ensure_origin_isolated)"] direction TB IsolateOrigin["_ensure_origin_isolated ━━━━━━━━━━ origin → file://clone_path upstream → real remote URL (if effective_url truthy)"] Decontaminate["decontaminate: ━━━━━━━━━━ git rm --cached GENERATED_FILES silent on failure (warns only)"] end T_ISOLATE_ERR(["RuntimeError origin isolation failed git remote add/set-url"]) T_SUCCESS_LOCAL(["★ CloneSuccessResult ━━━━━━━━━━ clone_source_type='local' remote_url=''"]) T_SUCCESS_REMOTE(["★ CloneSuccessResult ━━━━━━━━━━ clone_source_type='remote' remote_url=probed URL"]) %% ── PUSH RESILIENCE ──────────────────────────────────────────────────── subgraph PushPath ["● PUSH RESILIENCE (push_to_remote)"] direction TB ProtectedCheck{"protected branch?"} URLResolve{"● remote_url param provided? ━━━━━━━━━━ fall back to source_dir origin lookup"} BareCheck{"classify_remote_url == 'nonbare_local'?"} PushExec["● git push -u upstream {branch} ━━━━━━━━━━ 120 s timeout --force-with-lease if force=True"] end T_PUSH_PROT(["success=False error_type='protected_branch_push'"]) T_PUSH_NOURL(["success=False neither remote_url nor source_dir"]) T_PUSH_NONBARE(["success=False error_type='local_non_bare_remote'"]) T_PUSH_FAIL(["success=False stderr=git output error_type if force"]) T_PUSH_OK(["success=True"]) %% ── HOOK FORMATTER ───────────────────────────────────────────────────── subgraph FmtLayer ["★ HOOK OUTPUT FORMATTER (_fmt_clone_repo)"] FmtDetect{"★ payload keys?"} FmtWarn["★ ⚠ WARNING uncommitted_changes or unpublished_branch"] FmtFail["★ ✗ FAIL 'error' key present"] FmtOK["★ ✓ OK clone succeeded"] end %% ══ CONNECTIONS ════════════════════════════════════════════════════════ START -.->|"static analysis before any runtime call"| RuleGate RuleGate -->|"clone_local + remote_url capture"| RuleErr RuleGate -->|"templated/unknown + remote_url capture"| RuleWarn RuleErr --> T_RECIPE_BLOCKED RuleWarn --> T_RECIPE_BLOCKED RuleGate -->|"safe strategies / no capture"| SourceCheck SourceCheck -->|"missing"| T_VAL_ERR SourceCheck -->|"exists"| UncommittedGate UncommittedGate -->|"dirty (strategy='')"| T_UNCOMMITTED UncommittedGate -->|"clean / strategy≠''"| UnpublishedGate UnpublishedGate -->|"unpublished (exit 2)"| T_UNPUBLISHED UnpublishedGate -->|"ok / timeout / no origin"| Probe Probe -->|"returncode=0, url≠''"| ProbeOK Probe -->|"'no such remote' in stderr"| ProbeNoOrigin Probe -->|"≥30 s elapsed"| ProbeTimeout Probe -->|"other non-zero exit"| ProbeError ProbeOK --> StratDecide ProbeNoOrigin --> StratDecide ProbeTimeout --> StratDecide ProbeError --> StratDecide StratDecide -->|"strategy='clone_local'"| LocalCopy StratDecide -->|"proceed / default"| ProbeOKCheck ProbeOKCheck -->|"no (no_origin / timeout / error)"| T_PROBE_ERR ProbeOKCheck -->|"yes"| RemoteClone RemoteClone -->|"returncode≠0"| T_CLONE_FAIL RemoteClone -->|"success"| IsolateOrigin LocalCopy --> IsolateOrigin IsolateOrigin -->|"git error (not already-exists)"| T_ISOLATE_ERR IsolateOrigin -->|"ok"| Decontaminate Decontaminate -->|"clone_local path"| T_SUCCESS_LOCAL Decontaminate -->|"remote clone path"| T_SUCCESS_REMOTE T_SUCCESS_REMOTE -.->|"later in pipeline"| ProtectedCheck ProtectedCheck -->|"yes"| T_PUSH_PROT ProtectedCheck -->|"no"| URLResolve URLResolve -->|"neither provided"| T_PUSH_NOURL URLResolve -->|"resolved"| BareCheck BareCheck -->|"nonbare_local"| T_PUSH_NONBARE BareCheck -->|"network / bare_local / none"| PushExec PushExec -->|"returncode≠0"| T_PUSH_FAIL PushExec -->|"success"| T_PUSH_OK T_SUCCESS_LOCAL -.->|"result piped to hook"| FmtDetect T_SUCCESS_REMOTE -.->|"result piped to hook"| FmtDetect T_UNCOMMITTED -.->|"result piped to hook"| FmtDetect T_UNPUBLISHED -.->|"result piped to hook"| FmtDetect FmtDetect -->|"uncommitted_changes / unpublished_branch"| FmtWarn FmtDetect -->|"error key"| FmtFail FmtDetect -->|"neither"| FmtOK %% CLASS ASSIGNMENTS class START terminal; class RuleGate,RuleErr,RuleWarn newComponent; class T_RECIPE_BLOCKED,T_VAL_ERR,T_UNCOMMITTED,T_UNPUBLISHED,T_PROBE_ERR,T_CLONE_FAIL,T_ISOLATE_ERR terminal; class T_SUCCESS_LOCAL,T_SUCCESS_REMOTE,T_PUSH_OK terminal; class T_PUSH_PROT,T_PUSH_NOURL,T_PUSH_NONBARE,T_PUSH_FAIL terminal; class SourceCheck,UncommittedGate,UnpublishedGate detector; class Probe,ProbeOK,ProbeNoOrigin,ProbeTimeout,ProbeError newComponent; class StratDecide,ProbeOKCheck phase; class LocalCopy,RemoteClone handler; class IsolateOrigin,Decontaminate output; class ProtectedCheck,URLResolve,BareCheck detector; class PushExec handler; class FmtDetect,FmtWarn,FmtFail,FmtOK newComponent; ``` ### State Lifecycle Diagram ```mermaid %%{init: {'flowchart': {'nodeSpacing': 50, 'rankSpacing': 60, 'curve': 'basis'}}}%% flowchart TB %% CLASS DEFINITIONS %% classDef cli fill:#1a237e,stroke:#7986cb,stroke-width:2px,color:#fff; classDef stateNode fill:#004d40,stroke:#4db6ac,stroke-width:2px,color:#fff; classDef handler fill:#e65100,stroke:#ffb74d,stroke-width:2px,color:#fff; classDef phase fill:#6a1b9a,stroke:#ba68c8,stroke-width:2px,color:#fff; classDef newComponent fill:#2e7d32,stroke:#81c784,stroke-width:2px,color:#fff; classDef output fill:#00695c,stroke:#4db6ac,stroke-width:2px,color:#fff; classDef detector fill:#b71c1c,stroke:#ef5350,stroke-width:2px,color:#fff; classDef gap fill:#ff6f00,stroke:#ffa726,stroke-width:2px,color:#000; classDef terminal fill:#1a237e,stroke:#7986cb,stroke-width:2px,color:#fff; %% ENTRY %% CALL(["clone_repo(remote_url='', strategy=?)"]) subgraph RuntimeGates ["RUNTIME VALIDATION GATES — workspace/clone.py ●"] direction TB SRC_CHECK["G1 · source.is_dir() ━━━━━━━━━━ Raises ValueError if not a directory"] STRAT_CHECK{"strategy == '' ?"} UNCOMMIT["G2 · detect_uncommitted_changes ━━━━━━━━━━ git status --porcelain Returns gate result — no clone created"] UNPUB["G3 · detect_unpublished_branch ━━━━━━━━━━ git ls-remote --exit-code origin Returns gate result — no clone created"] PROBE["G4 · _probe_origin_url ● ━━━━━━━━━━ Raises RuntimeError if reason != ok NO silent local-path fallback"] end subgraph GateResults ["GATE RESULT TYPES — core/_type_results.py ●"] direction LR RES_UNCOMMIT["CloneGateUncommitted ━━━━━━━━━━ uncommitted_changes: Literal['true'] changed_files: str total_changed: str branch, source_dir — NO clone_path — NO remote_url —"] RES_UNPUB["CloneGateUnpublished ━━━━━━━━━━ unpublished_branch: Literal['true'] branch, source_dir — NO clone_path — NO remote_url —"] end subgraph RemoteResolution ["remote_url RESOLUTION — workspace/clone.py ●"] direction TB INPUT_URL["Input: remote_url = '' (optional) ━━━━━━━━━━ Caller may supply or leave blank"] LOCAL_STRAT{"strategy == 'clone_local' ?"} USE_PROBE["effective_url = resolution.url ━━━━━━━━━━ Real remote (from G4 probe)"] USE_CALLER["effective_url = caller_url OR resolution.url ━━━━━━━━━━ Caller override takes priority"] ISOLATE["_ensure_origin_isolated ━━━━━━━━━━ origin → file://clone_path (unique) upstream → effective_url"] end subgraph SuccessResult ["SUCCESS CONTRACT — core/_type_results.py ●"] direction TB CLONE_PATH["clone_path: str [INIT_ONLY] ━━━━━━━━━━ <run_name>-YYYYMMDD-HHMMSS-ffffff Unique per call — never reused"] SRC_DIR["source_dir: str [INIT_ONLY] ━━━━━━━━━━ expanduser().resolve() — fixed at entry"] REMOTE_URL["remote_url: str [RESOLVED · required] ━━━━━━━━━━ Non-Optional on CloneSuccessResult '' for clone_local · real URL for proceed"] SRC_TYPE["clone_source_type: Literal['remote','local'] [INIT_ONLY] ━━━━━━━━━━ 'local' when strategy='clone_local'"] SRC_REASON["clone_source_reason: str [INIT_ONLY] ━━━━━━━━━━ 'strategy_clone_local' or 'ok'"] end subgraph RecipeGate ["★ RECIPE SEMANTIC GATE — recipe/rules_clone.py ●"] direction TB RULE["clone-local-strategy-with-remote-url-capture ━━━━━━━━━━ Fires at recipe-load time (open_kitchen / load_recipe) Checks: python=clone_repo AND capture references result.remote_url Inspects template VALUE not capture key name"] STRAT_EVAL{"strategy value?"} PASS_RULE["PASS — strategy='' / 'proceed' ━━━━━━━━━━ Part A guarantees non-empty remote_url Safe to capture downstream"] ERROR_RULE["ERROR — strategy='clone_local' literal ━━━━━━━━━━ remote_url always '' for local transport Capturing it corrupts downstream push"] WARN_RULE["WARN — strategy=${{template}} ━━━━━━━━━━ Cannot determine statically May produce empty remote_url"] end subgraph Formatter ["★ DISPLAY LAYER — hooks/_fmt_status.py (new tests)"] FMT["_fmt_clone_repo ━━━━━━━━━━ All scalar fields rendered unconditionally remote_url: '' rendered as-is for local clones Sentinel key PRESENCE → WARNING mark (not value)"] end %% RUNTIME FLOW %% CALL --> SRC_CHECK SRC_CHECK -->|invalid dir| ERR_VAL(["ValueError"]) SRC_CHECK -->|valid| STRAT_CHECK STRAT_CHECK -->|yes · check clean| UNCOMMIT UNCOMMIT -->|changes found| RES_UNCOMMIT UNCOMMIT -->|clean| UNPUB UNPUB -->|branch absent| RES_UNPUB UNPUB -->|published| PROBE STRAT_CHECK -->|no| PROBE PROBE -->|reason != ok| ERR_PROBE(["RuntimeError no_origin"]) PROBE -->|reason == ok| INPUT_URL %% RESOLUTION FLOW %% INPUT_URL --> LOCAL_STRAT LOCAL_STRAT -->|yes — shutil.copytree| USE_CALLER LOCAL_STRAT -->|no — git clone| USE_PROBE USE_PROBE --> USE_CALLER USE_CALLER --> ISOLATE ISOLATE --> CLONE_PATH ISOLATE --> SRC_DIR ISOLATE --> REMOTE_URL ISOLATE --> SRC_TYPE ISOLATE --> SRC_REASON %% DISPLAY FLOW %% CLONE_PATH --> FMT SRC_DIR --> FMT REMOTE_URL --> FMT SRC_TYPE --> FMT SRC_REASON --> FMT RES_UNCOMMIT --> FMT RES_UNPUB --> FMT %% RECIPE GATE FLOW %% RULE --> STRAT_EVAL STRAT_EVAL -->|safe strategy| PASS_RULE STRAT_EVAL -->|clone_local literal| ERROR_RULE STRAT_EVAL -->|template expr| WARN_RULE %% CLASS ASSIGNMENTS %% class CALL terminal; class SRC_CHECK,UNCOMMIT,UNPUB,PROBE detector; class STRAT_CHECK,LOCAL_STRAT,STRAT_EVAL phase; class RES_UNCOMMIT,RES_UNPUB gap; class INPUT_URL,USE_PROBE,USE_CALLER,ISOLATE handler; class CLONE_PATH,SRC_DIR,SRC_TYPE,SRC_REASON stateNode; class REMOTE_URL gap; class RULE detector; class PASS_RULE output; class ERROR_RULE detector; class WARN_RULE gap; class FMT output; class ERR_VAL,ERR_PROBE terminal; ``` ### Process Flow Diagram ```mermaid %%{init: {'flowchart': {'nodeSpacing': 40, 'rankSpacing': 50, 'curve': 'basis'}}}%% flowchart TB %% CLASS DEFINITIONS %% classDef terminal fill:#1a237e,stroke:#7986cb,stroke-width:2px,color:#fff; classDef stateNode fill:#004d40,stroke:#4db6ac,stroke-width:2px,color:#fff; classDef handler fill:#e65100,stroke:#ffb74d,stroke-width:2px,color:#fff; classDef phase fill:#6a1b9a,stroke:#ba68c8,stroke-width:2px,color:#fff; classDef detector fill:#b71c1c,stroke:#ef5350,stroke-width:2px,color:#fff; classDef output fill:#00695c,stroke:#4db6ac,stroke-width:2px,color:#fff; classDef newComponent fill:#2e7d32,stroke:#81c784,stroke-width:2px,color:#fff; %% TERMINALS %% START([START]) GATED([GATED — early return]) FAIL([ERROR — raises RuntimeError]) COMPLETE([COMPLETE]) subgraph Entry ["Entry & Resolution"] direction TB A["● clone_repo ━━━━━━━━━━ workspace/clone.py args: source_dir, run_name, branch, strategy, remote_url"] B["Resolve source_dir ━━━━━━━━━━ detect_source_dir(cwd) if source_dir empty"] C["Detect branch ━━━━━━━━━━ detect_branch(source_dir) blank if HEAD / detached"] end subgraph Gates ["Strategy Gates — bypassed when strategy='clone_local' or 'proceed'"] direction TB D{"strategy bypasses gates?"} E["detect_uncommitted_changes ━━━━━━━━━━ git status --porcelain reads: working tree"] F{"any uncommitted changes?"} G["detect_unpublished_branch ━━━━━━━━━━ git ls-remote origin 15 s timeout — fail-open"] H{"branch unpublished?"} GU["CloneGateUncommitted ━━━━━━━━━━ writes: {uncommitted_changes, changed_files, total_changed}"] GP["CloneGateUnpublished ━━━━━━━━━━ writes: {unpublished_branch, branch, source_dir}"] end subgraph Dispatch ["Clone Dispatch"] direction TB K{"strategy == 'clone_local'?"} L["● _probe_origin_url ━━━━━━━━━━ git remote get-url origin 30 s timeout reads: git remote config writes: CloneSourceResolution"] M{"resolution.reason == 'ok'?"} N["shutil.copytree ━━━━━━━━━━ full worktree copy source_type = 'local' reason = 'strategy_clone_local'"] O["git clone --branch ━━━━━━━━━━ resolution.url → clone_path 300 s timeout source_type = 'remote'"] end subgraph Isolation ["● Origin Isolation Fix — #807 / #377"] direction TB P["● _ensure_origin_isolated ━━━━━━━━━━ UNCONDITIONAL on both paths writes: origin = file://clone_path (self-referential — breaks repo aliasing)"] Q{"effective_url truthy?"} R["_add_or_set_upstream ━━━━━━━━━━ writes: upstream = real remote URL preserves provenance for push_to_remote"] S["Decontaminate clone ━━━━━━━━━━ git rm --cached GENERATED_FILES git commit --no-verify os.unlink disk copies"] end subgraph Result ["● Result Assembly"] direction TB T["● CloneSuccessResult ━━━━━━━━━━ clone_path, source_dir remote_url (effective_url) ★ clone_source_type: 'remote'|'local' ★ clone_source_reason: str writes: _type_results.py"] end subgraph Downstream ["Downstream Consumers"] direction LR U["★ _fmt_clone_repo ━━━━━━━━━━ hooks/_fmt_status.py renders clone_source_type + clone_source_reason as flat KV output"] V["● Recipe Validator ━━━━━━━━━━ run_semantic_rules() reads: Recipe steps + captures"] end subgraph RuleLogic ["● New Rule: clone-local-strategy-with-remote-url-capture"] direction TB W{"step = clone_repo AND captures result.remote_url?"} X{"strategy value?"} RE["★ RuleFinding ERROR ━━━━━━━━━━ strategy='clone_local' remote_url always ''\n→ recipe INVALID fixture: invalid.yaml"] RW["★ RuleFinding WARNING ━━━━━━━━━━ template / unknown strategy cannot resolve statically → recipe VALID, flagged"] RN["No Finding ━━━━━━━━━━ safe/proceed/empty strategy or no remote_url capture fixture: valid.yaml"] end %% MAIN FLOW %% START --> A A --> B --> C C --> D D -->|"yes — skip gates"| K D -->|"no"| E E --> F F -->|"yes"| GU --> GATED F -->|"no"| G G --> H H -->|"yes"| GP --> GATED H -->|"no"| K K -->|"yes (clone_local)"| N K -->|"no"| L L --> M M -->|"no — no usable origin"| FAIL M -->|"yes"| O N --> P O --> P P --> Q Q -->|"yes"| R --> S Q -->|"no"| S S --> T T --> U T --> V V --> W W -->|"no"| RN W -->|"yes"| X X -->|"'clone_local'"| RE X -->|"template / unknown"| RW X -->|"'' / 'proceed'"| RN RE --> COMPLETE RW --> COMPLETE RN --> COMPLETE %% CLASS ASSIGNMENTS %% class START,GATED,FAIL,COMPLETE terminal; class A,N,O,P,R,S handler; class B,C,L phase; class D,F,H,K,M,Q,W,X stateNode; class GU,GP detector; class T output; class U newComponent; class V,RE,RW detector; class RN phase; ``` Closes #807 ## Implementation Plan Plan file: `/home/talon/projects/autoskillit-runs/remediation-807-20260413-075530-622615/.autoskillit/temp/rectify/rectify_clone_repo_silent_fallback_2026-04-13_110616_part_a.md` 🤖 Generated with [Claude Code](https://claude.com/claude-code) via AutoSkillit ## Token Usage Summary | Step | uncached | output | cache_read | cache_write | count | time | |------|----------|--------|------------|-------------|-------|------| | investigate | 120 | 14.5k | 477.8k | 78.4k | 1 | 5m 38s | | rectify | 4.1k | 68.6k | 2.2M | 233.2k | 2 | 25m 4s | | review | 2.9k | 8.5k | 226.5k | 55.4k | 1 | 6m 55s | | dry_walkthrough | 3.2k | 45.6k | 2.7M | 137.3k | 2 | 15m 50s | | implement | 2.4k | 49.1k | 7.0M | 146.4k | 2 | 19m 12s | | assess | 1.2k | 25.3k | 2.6M | 77.2k | 1 | 11m 51s | | prepare_pr | 60 | 6.7k | 212.6k | 30.7k | 1 | 1m 43s | | run_arch_lenses | 3.2k | 31.5k | 622.1k | 118.9k | 3 | 9m 49s | | compose_pr | 67 | 11.6k | 243.5k | 38.1k | 1 | 2m 46s | | **Total** | 17.2k | 261.3k | 16.3M | 915.6k | | 1h 38m | --------- Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

Trecek commented Apr 11, 2026

View reviewed changes

Trecek added this pull request to the merge queue Apr 12, 2026

github-merge-queue bot removed this pull request from the merge queue due to failed status checks Apr 12, 2026

Trecek added this pull request to the merge queue Apr 12, 2026

github-merge-queue bot removed this pull request from the merge queue due to failed status checks Apr 12, 2026

Trecek and others added 6 commits April 11, 2026 22:27

fix: update sentinel tests, contract assertions, and line limits for …

521b099

…IDLE_STALL

fix: update list-payload site line numbers for smoke_utils.py

c38b460

fix(review): cancel dangling io_task on timeout in remote_resolver

7e1044e

Add io_task.cancel() in the TimeoutError handler to prevent a task leak when proc.communicate() future is abandoned after timeout. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Trecek force-pushed the impl-rectify-idle-stall-watchdog-20260411-142908 branch from deeff77 to 4584ee1 Compare April 12, 2026 05:30

Trecek enabled auto-merge April 12, 2026 05:32

Trecek added this pull request to the merge queue Apr 12, 2026

Merged via the queue into integration with commit 225d904 Apr 12, 2026
2 checks passed

Trecek deleted the impl-rectify-idle-stall-watchdog-20260411-142908 branch April 12, 2026 05:39

Trecek mentioned this pull request Apr 14, 2026

Promote integration to main (198 PRs, 182 issues, 102 fixes, 107 features, 1 infra, 7 tests, 4 docs) #925

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Rectify: Stdout Idle Watchdog + Bounded Suppression for Stall Detection#735

Rectify: Stdout Idle Watchdog + Bounded Suppression for Stall Detection#735
Trecek merged 6 commits intointegrationfrom
impl-rectify-idle-stall-watchdog-20260411-142908

Trecek commented Apr 11, 2026

Uh oh!

Trecek left a comment

Uh oh!

Uh oh!

Trecek Apr 11, 2026

Uh oh!

Trecek Apr 11, 2026

Uh oh!

Uh oh!

Trecek Apr 11, 2026

Uh oh!

Trecek Apr 11, 2026

Uh oh!

Trecek Apr 11, 2026

Uh oh!

Trecek Apr 11, 2026

Uh oh!

Trecek Apr 11, 2026

Uh oh!

Trecek Apr 11, 2026

Uh oh!

Trecek Apr 11, 2026

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

Trecek commented Apr 11, 2026

Summary

Changed Files

New (★)

Modified (●)

Test Plan

Implementation Plan

Uh oh!

Trecek left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant