* fix(worker): port keygraph shared-core correctness fixes
* refactor(worker): adopt collectors/ and ai/pi/ layout; add task budget cap and cancellation
* refactor(worker): drop inconsistent Collector "Server" suffix
* refactor(worker): drop unused providerConfig/apiKey seams, resolve credentials from env only
* refactor(worker): port oss code_path pattern expansion + external_directory allow
* fix(worker): preserve dotfile paths in code_path avoid patterns (.env no longer stripped to env)
* feat(worker): render Unprocessed Vulnerabilities section in exploit deliverable (align with oss)
* feat(worker): request set_blind_spots for all vuln classes (align auth/ssrf with production prompts)
* refactor(worker): adopt unified permissionSystem* naming and helper layout
* refactor(worker): inline blind_spots into vuln deliverable section array
* chore(worker): drop unused zod dependency (tree is typebox-native)
* fix(worker): normalize base32 TOTP secret to accept padding and whitespace
* refactor(worker): adopt shared toolResult helper and flatSchema naming in collectors
* refactor(worker): use undefined over null in queue-schema builders
* docs(worker): converge renderer/collector doc comments to current pi terminology
* refactor(worker): adopt schema.ts cleanInput/stringEnum helpers in collectors
* feat(worker): converge exploit-collector/renderer with vendored; capture and render overview for blocked findings
* refactor(worker): converge session-tools/pipeline/exploitation-checker with vendored
* refactor(worker): converge task-tool usage reporting with vendored onUsage callback
* refactor(worker): converge structured output onto a submitTool executor channel
* docs(worker): expand exploit-renderer docstring to match shannon-oss
* docs(worker): adopt richer vuln-renderer docstring from shannon-oss
* docs(worker): neutralize billing-detection wording for shannon-oss parity
* fix(worker): verify checkpoint hash in the deliverables clone being reset
* fix(worker): fail fast on malformed exploitation queue JSON
* fix(worker): honor retryable flag when classifying exploitation-queue check failures
* fix(worker): fail fast on corrupted session.json in run-scope validation
* feat(worker): propagate Temporal cancellation signal into agent and auth pi sessions
* fix(worker): mark exploit agent complete when exploitation is skipped so resume skips it
* prompts: drop scan description from executive report prompt
* refactor(worker): add createGenericSubmitTool for raw JSON-schema submit tools
* refactor(worker): gate playwright-cli skill to browser agents via skillsOverride (adopt shannon-oss mechanism)
* docs(worker): correct formatLogTime comment to UTC to match toISOString
* refactor(worker): converge queue-schemas with shannon-oss (guarded count, decl order)
* refactor(worker): converge task-tool with shannon-oss (byte-identical; modelRegistry optional)
* fix(worker): use replaceLiteral for all prompt value insertions to prevent $-mangling
* fix(worker): classify agent execution failures by error type instead of hardcoding validation
* fix(worker): cap auth-failure detail at 250 chars to match shannon-oss
* style(worker): apply biome formatting
* refactor(worker): remove per-session task delegation cap from task tool
* feat(steerability): add config-driven profile with code_path avoid enforcement
* fix(steerability): write SDK deny rules once per workflow to avoid parallel-agent race
* fix(steerability): reference guidance by pointer in report DROP rules
* fix(steerability): tighten code_path avoid enforcement
* chore(steerability): use shared ALL_VULN_CLASSES const and tighten RunScope type
* fix(steerability): validate run scope before resume short-circuit
* fix(steerability): emit only documented Read/Edit deny rules for code_path
* fix(steerability): assemble report from analysis deliverables when exploit is disabled
* feat(steerability): preflight check that code_path rules match at least one repo entry
* fix(steerability): tag missing code_path entries with avoid/focus kind
* revert(steerability): assemble report from analysis deliverables when exploit is disabled
* feat(steerability): render per-class findings from queue JSON when exploit is disabled
* refactor(steerability): trim findings renderer to common mappable rows
* feat(steerability): allow report agent to rewrite category-label finding titles
* docs(steerability): document new config fields in README and CLAUDE.md
* docs(steerability): comment out optional config sections in examples
* feat: add structured outputs for vuln agent exploitation queues
Use Claude Agent SDK's native outputFormat to get schema-validated JSON
queue data from vulnerability analysis agents instead of relying on
save-deliverable tool calls for queue files.
- Add Zod schemas for all 5 vuln types (injection, xss, auth, ssrf, authz)
- Thread outputFormat through SDK call chain (executor → message handlers)
- Write structured_output to disk as queue JSON before validation
- Handle error_max_structured_output_retries as retryable failure
- Update vuln prompts to use structured output for queues
- Keep save-deliverable for markdown deliverables (unchanged)
* fix: correct structured output schema conversion for Claude Agent SDK
Use draft-07 target for z.toJSONSchema() instead of the default
draft-2020-12, which the SDK's AJV validator doesn't support. Update
pipeline-testing prompts to use structured output instead of raw JSON
responses.
* refactor: remove save-deliverable references for queues in vuln prompts
Queues are now captured via structured outputs, so vuln agents no longer
need to use save-deliverable for queue JSON. Removes references to
"structured response/output" phrasing and aligns all prompts to use
consistent "exploitation queue" terminology.
* refactor: remove queue support from save-deliverable
Queues are now produced via structured outputs, so save-deliverable no
longer needs queue-related code. Removes queue enum values, filename
mappings, JSON validation, and updates all prompt tool descriptions to
match the simplified CLI interface.
* fix: instruct vuln agents to save deliverable before exploitation queue
The structured output tool terminates the agent session when called.
Agents were calling it before saving their deliverable markdown,
causing output validation failures and unnecessary retries.
* refactor: remove explicit exploitation queue output instructions from vuln prompts
The Claude Agent SDK automatically captures structured output on the
last turn when outputFormat is set. Prompts explicitly telling agents
to produce the queue caused them to call StructuredOutput mid-session,
conflicting with the SDK mechanism and silently dropping the output.
Removed exploitation_queue_requirements sections and queue references
from conclusion triggers. Added note that the queue is captured
automatically. Updated Your Output to point to the deliverable markdown.
