Commit Graph
7 Commits
Author SHA1 Message Date
Renn F 1d02b09fe0 fix(bash-guard): deny interpreter/library HTTP to internal hosts (#175)
The internal-API rule only fired when the FIRST shell token was an HTTP
CLI (curl/wget/http/https/httpie). smoke-17 showed an agent reach the
orchestrator with hand-forged X-Agent-ID/X-Agent-Role headers via:

  python3 << 'EOF'
  import httpx
  httpx.post("http://roboco-orchestrator:8000/api/v2/flow/developer/i_will_work_on",
             headers={"X-Agent-ID": "<self>", "X-Agent-Role": "developer"})
  EOF

The binary is python3 (slips the CLI check) and it imports httpx, not
roboco.* (slips the #164 import check). Only minimax's wrong endpoint
path prevented a real gateway bypass under a forged identity.

Add a language-agnostic rule: deny when the command pairs an HTTP-client
token (httpx/requests/urllib/aiohttp/http.client/net::http/fetch(/
node-fetch/axios/...) with a forbidden internal host, consistent with
the curl/wget sibling (inspects full $low incl. heredoc body). External
HTTP (pypi/docs/github) has no internal host so it still passes. The
stale "interpreter one-liners — out of scope" KNOWN GAP comment is
corrected; the variable-expansion gap remains documented.

11 new tests incl. the exact smoke-17 heredoc, requests/urllib/aiohttp/
node-fetch/Net::HTTP variants, and allow-cases (external host, client
import w/o host, pytest runner). make quality green.
2026-05-18 00:08:04 +02:00
Renn F 1605d187f1 fix(security): bash-guard git-ops check inspects commands, not file content (#165)
The git network/auth deny rule matched its regex against the whole
command string, so heredoc bodies and echo/printf arguments that merely
documented git verbs (a README, a notes file) were treated as git
invocations and denied. This wedged smoke-13's dev: after wiping the
README via an Edit/Write fallback it could not restore it because every
`cat > README.md << EOF ... git commit ... EOF` was blocked.

The git-ops check now runs against a skeleton of the command with
heredoc bodies and echo/printf literal args stripped (both are data the
shell writes, never executed). Quoted args to a shell interpreter
(`bash -c "... && git fetch"`) ARE executed, are not echo/printf/heredoc
bodies, and so survive untouched — the hook's core purpose is preserved.
A sentinel prefix distinguishes a legitimately-empty skeleton from a
python failure (fail closed on failure). All other rules, including the
#164 import-bypass rule, still inspect the full command.
2026-05-16 02:20:08 +02:00
Renn F 81f5655d48 fix(security): block gateway-internals import + agent-id forgery (#164)
Smoke-12: be-dev-1 (minimax-m2.7) bypassed the entire MCP boundary by
running `uv run python3 -c "import os;
os.environ['ROBOCO_AGENT_ID']='...'; from roboco.mcp.flow_server
import open_pr; open_pr(...)"` from the Bash tool. This voided the
per-role tool manifest (role-scoping is meaningless if the agent can
import any server module in-process), forged agent identity via an
env-var rewrite, and ran choreographer code outside the gateway's
tracing + auth.

bash-guard-hook.sh now adds two deny rules:
  1. Any python/uv/poetry/pipenv/pdm/hatch invocation that imports or
     `-m`-runs roboco.* internals (mcp/services/runtime/foundation/
     api/enforcement). The whole command string — heredoc body
     included — is matched, so quoting/heredoc forms are covered.
  2. Any assignment or export of ROBOCO_AGENT_ID (identity forgery).

Reading roboco source for context (cat/grep) is still allowed — the
block is on *executing* internals, not viewing them. Normal python
one-liners without roboco imports still pass.

19 bash-guard tests pass (10 prior + 9 new). Note: takes effect on
agent-image rebuild (hook ships in the agent container).
2026-05-16 00:59:26 +02:00
Renn F 4829f93a68 fix(gateway): unblock task claim; full Phase 0/1/2 remediation
Resolves the 100% claim-failure rate introduced by the gateway rewrite
  (commit 62bda0c plus 78 follow-ups). Live smoke runs hit
  `404 /api/v2/flow/developer/...` on every dev verb plus a manifest
  fallback that silently exposed off-role verbs to PMs — confirmed
  firing simultaneously in NAS agent logs (be-dev-1, be-pm, main-pm).

  Audit reports under docs/internal/audit_2026_05_04/ catalogue 49
  defects across gateway, services, prompts, MCP transport, substrate,
  and tests (8 detail reports + master synthesis). Six smoking guns;
  three proven in production logs.

  Phase 0 — unblock claim:
  - URL prefix /api/v2/flow/dev → /developer; slug-map board roles
    (product_owner, head_marketing) → /board (D-01)
  - _i_will_work_on AttributeError on None across pending /
    needs_revision / claimed re-entry branches (D-02)
  - Seed last_heartbeat_at in _qa_or_doc_claim (D-03)
  - Drop misleading i_have_committed verb; dev flow uses commit() (D-04)
  - Manifest mount via compose; flow_server + do_server fail loud
    instead of exposing all-verbs fallback (D-12)
  - MCP _post() surfaces envelope body on 4xx so agents see remediate
    hints (D-13)
    on git failure so retries aren't blocked by half-state (S-01)

  Phase 1 — lifecycle stability:
  - _resolve_skill falls back to AgentTable.capabilities (D-06)
  - main_pm_complete uses kwargs for escalate_to_ceo (D-07)
  - i_am_done auto-runs submit_verification when in_progress (D-08)
  - active_claimant_id wired in claim/unclaim paths — single-claimant
    invariant now functional (D-05)
  - qa_pass/qa_fail assert claimed_by parity with qa_agent_id (D-18)
  - Prompt-drift sweep: fail() shape, i_am_done(task_id, notes),
    subtask cap (12 hard / 8 soft), error-code symbology rewritten in
    base.md + per-role anti-patterns (D-10/11/29/30/31, D-37)

  Phase 2 — invariants + architecture:
  - Real-DB integration test exercising claim → in_progress → commit
    → submit_for_qa → i_am_done → awaiting_qa (P2-1)
  - choreographer.py → package; 3 of 6 role mixins extracted
    (board, doc, qa). _impl.py 2,526 → 2,080 lines (-18%). Continuation
    plan in docs/internal/audit_2026_05_04/p2_2_decompose_plan.md (P2-2)
  - Closure guards consolidated via _subtasks_not_terminal_envelope (P2-3)
  - TaskService.unclaim_for_reaper routed through canonical
    _validate_and_set_status; in_progress → pending added to
    VALID_TRANSITIONS (P2-4)
  - Dead code removed: i_am_done_with_catchup verb, _run_catch_up helper
    (P2-5)
  - 6 state-machine invariants asserted via property test (P2-6)
  - attempt_id (uuid4) stamped on every gateway.rejected audit row (P2-7)
  - _reconcile_orphan_claims_on_startup rolls back tasks left CLAIMED
    with branch_name=NULL from prior crashes (P2-8)
  - scripts/regenerate_verb_tables.py introspects Pydantic schemas +
    role_config; compose_prompt injects per-role tables as a layer.
    Eliminates the prompt-drift class structurally (P2-9)

  Other:
  - D-48: orchestrator mounts host's ~/.claude.json when present so
    agents don't boot from backup recovery on every spawn
  - D-49: dev dispatcher rejects role-mismatched spawns (e.g. doc task
    assigned to dev agent)

  Tests: 553 pass · ruff + mypy clean. Live NAS smoke verification
  pending — needs the stack brought back up.
2026-05-04 23:43:55 +02:00
Renn F b0107f8c12 fix(optimal): raise instead of falling back to 'unknown' doc source
Silent fallback hid an upstream bug where journal_entry.id wasn't
flushed before indexing. Raise so we see the regression.
2026-05-03 10:31:59 +02:00
Renn F c0c5838baa fix(bash-guard): close scheme-less curl bypass
Previous regex used a greedy [^|]* and required at least one / before
the host, so 'curl roboco-orchestrator:8000/api' (no scheme, no slash)
slipped through. Split into two simpler checks: (a) line starts with
curl/wget/http/https/httpie, AND (b) line contains a forbidden host.
Probe-verified: scheme-ful, scheme-less, and protocol-relative forms
all denied; external URLs still allowed; GitHub-specific deny still
fires first.
2026-05-03 06:17:37 +02:00
Renn F 8381ade3ce fix(bash-guard): deny internal curl to orchestrator/localhost
Prompts told agents internal API calls were denied; the guard only
denied GitHub. Combined with task 4 (X-Agent-Role enforcement) this
closes the manifest-bypass loophole.
2026-05-03 06:11:38 +02:00