feat: add CI operational checkpoint v0.1

.github/workflows/agent-continuity-conformance.yml

@@ -0,0 +1,41 @@
+name: Agent Continuity conformance
+
+on:
+  pull_request:
+    paths:
+      - "standards/agent-continuity/**"
+      - ".github/workflows/agent-continuity-conformance.yml"
+  push:
+    paths:
+      - "standards/agent-continuity/**"
+      - ".github/workflows/agent-continuity-conformance.yml"
+  workflow_dispatch:
+
+permissions:
+  contents: read
+
+jobs:
+  conformance:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Check out repository
+        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
+        with:
+          persist-credentials: false
+
+      - name: Set up Python
+        uses: actions/setup-python@a26af69be951a213d495a4c3e4e4022e16d87065 # v5.6.0
+        with:
+          python-version: "3.12"
+
+      - name: Install conformance dependency
+        run: |
+          python -m pip install --disable-pip-version-check \
+            -r standards/agent-continuity/conformance/requirements.txt
+
+      - name: Run Agent Continuity conformance suite
+        run: |
+          python -m unittest discover \
+            -s standards/agent-continuity/conformance \
+            -p 'test_*.py' \
+            -v

standards/agent-continuity/CI-CHECKPOINT-CODES.md

@@ -0,0 +1,29 @@
+# CI Operational Checkpoint v0.1 Outcome Registry
+
+Consumers MUST branch on `outcome` and `reason_code`, not on `detail`.
+
+## Outcomes
+
+| Outcome | Meaning |
+|---|---|
+| `CONTINUE` | The checkpoint is valid, lineage is coherent, the workspace matches, and the next step may resume as context-only work. |
+| `REVALIDATE_WORKSPACE` | Repository identity matches, but required workspace evidence is missing or base, head, or dirty-state evidence changed. |
+| `RESTART_REQUIRED` | The checkpoint cannot safely resume in the current workspace or failed schema/digest validation. |
+| `IDEMPOTENT_REPLAY` | The same checkpoint was already consumed; no duplicate work should be created. |
+| `REJECT_LINEAGE_MISMATCH` | Parent, checkpoint identity, time, objective, constraints, sequence, trajectory, rejected-approach continuity, or parent integrity is inconsistent. |
+| `REJECT_UNVERIFIED_COMPLETION` | Verification was completed without durable evidence, disappeared, or conflicts with prior proof or pending-work lineage. |
+| `REJECT_INVALID_AUTHORITY` | Continuity material attempted to carry action authority or bypass fresh authorization. |
+
+## Reason codes
+
+| Outcome | Reason code |
+|---|---|
+| `CONTINUE` | `CONTINUE_OK` |
+| `REVALIDATE_WORKSPACE` | `WORKSPACE_STATE_CHANGED`, `CURRENT_WORKSPACE_FIELD_MISSING` |
+| `RESTART_REQUIRED` | `SCHEMA_INVALID`, `DIGEST_MISMATCH`, `WORKSPACE_IDENTITY_MISMATCH`, `CURRENT_WORKSPACE_FIELD_MISSING` |
+| `IDEMPOTENT_REPLAY` | `CHECKPOINT_ALREADY_CONSUMED` |
+| `REJECT_LINEAGE_MISMATCH` | `ROOT_HAS_PARENT`, `PARENT_REQUIRED`, `PARENT_NOT_FOUND`, `PREVIOUS_CHECKPOINT_REQUIRED`, `PREVIOUS_CHECKPOINT_SCHEMA_INVALID`, `PREVIOUS_CHECKPOINT_DIGEST_MISMATCH`, `PREVIOUS_CHECKPOINT_SEMANTIC_INVALID`, `CHECKPOINT_ID_REUSED`, `PARENT_MISMATCH`, `SEQUENCE_MISMATCH`, `TRAJECTORY_CHANGED`, `CREATION_TIME_REGRESSED`, `OBJECTIVE_CHANGED`, `CONSTRAINT_LOST`, `REJECTED_APPROACH_ID_DUPLICATED`, `REJECTED_APPROACH_LOST`, `REJECTED_APPROACH_CHANGED` |
+| `REJECT_UNVERIFIED_COMPLETION` | `VERIFICATION_ID_DUPLICATED`, `VERIFICATION_SET_MISMATCH`, `COMPLETION_EVIDENCE_MISSING`, `MEMORY_IS_NOT_VERIFICATION`, `COMPLETED_VERIFICATION_LOST`, `COMPLETED_VERIFICATION_CHANGED`, `PENDING_VERIFICATION_LOST`, `PENDING_VERIFICATION_CHANGED` |
+| `REJECT_INVALID_AUTHORITY` | `AUTHORITY_NOT_CONTEXT_ONLY`, `FRESH_AUTHORITY_REQUIRED` |
+
+Within version `0.1`, existing reason codes MUST NOT change meaning.

standards/agent-continuity/CI-OPERATIONAL-CHECKPOINT-V0.1.md

@@ -0,0 +1,107 @@
+# CI Operational Checkpoint v0.1
+
+- **Status:** Draft
+- **Profile:** `pythia.ci_operational_checkpoint.v0.1`
+- **Parent:** [RFC-001](RFC-001-VERIFIABLE-CONTINUATION-ENVELOPE.md)
+- **Scope:** CI/CD phase transitions, retries, restart, and agent handoff
+
+## Purpose
+
+This profile defines a bounded checkpoint for resuming delivery work without
+reconstructing state from narrative memory and without inheriting authority for
+a consequential action.
+
+```text
+validate schema and digest
+  -> check replay
+  -> validate parent checkpoint and lineage
+  -> preserve objective, constraints, rejected rationale, and verification proof
+  -> compare current workspace
+  -> CONTINUE | REVALIDATE | RESTART | IDEMPOTENT_REPLAY | REJECT
+```
+
+## Core boundary
+
+> Context may continue. Authority must be evaluated again.
+
+`authority` is fixed to `context_only`. Merge and deploy intent MUST set
+`requires_fresh_authority=true`. A checkpoint never replaces an Action Envelope,
+current checks, current reviews, or an executor guard.
+
+## Required content
+
+A conforming checkpoint records trajectory and parent identity, source agent and
+session, repository and workspace identity, objective, phase, constraints,
+rejected approaches, touched resources, verification state, the next action,
+and a canonical SHA-256 digest. `dirty_state_digest` is optional.
+
+The strict schema is
+[`schema/ci-operational-checkpoint-v0.1.schema.json`](schema/ci-operational-checkpoint-v0.1.schema.json).
+
+## Active-state continuity
+
+Within one trajectory:
+
+- child creation time MUST NOT precede parent creation time;
+- objective and acceptance criteria MUST remain unchanged;
+- every prior `must` and `must_not` constraint MUST remain active;
+- every rejected approach row and rationale MUST remain unchanged;
+- every completed verification target and prior evidence reference MUST remain;
+- every pending verification MUST remain with the same target or move to
+  completed with durable evidence.
+
+Constraints and evidence may be added. Existing active state and proof may not be
+silently removed or rewritten. A changed objective requires a new trajectory.
+
+## Verification boundary
+
+Completed verification requires durable evidence. Memory and agent summaries
+are not verification evidence, regardless of URI-scheme letter case. Additional
+evidence may be appended, but existing evidence cannot be removed or replaced.
+
+## Parent-checkpoint integrity
+
+Every non-root resume requires the full previous checkpoint. A known parent ID
+alone is insufficient because it cannot prove active-state or verification
+continuity.
+
+Before comparison, the previous checkpoint MUST pass schema validation,
+canonical digest verification, identifier uniqueness, verification-set
+consistency, and the memory-evidence boundary. Invalid or missing parent material
+returns `REJECT_LINEAGE_MISMATCH`.
+
+Durable lookup and storage remain follow-up work in issue #223.
+
+## Workspace comparison
+
+A repository or working-directory mismatch returns `RESTART_REQUIRED`. Changes
+to base ref, head SHA, or a declared dirty-state digest return
+`REVALIDATE_WORKSPACE`.
+
+When the checkpoint declares `dirty_state_digest`, the resumed runtime MUST
+explicitly report it, including `null` for an observed clean workspace. Missing
+observation returns `CURRENT_WORKSPACE_FIELD_MISSING`. If the checkpoint omits
+the field, it is outside the comparison boundary.
+
+## Replay
+
+A consumed checkpoint ID returns `IDEMPOTENT_REPLAY`; completed work must not be
+repeated.
+
+## Relationship to Action Envelope V1
+
+The checkpoint says what context and verification state to restore. Action
+Envelope V1 decides whether an exact consequential action may execute now. The
+contracts are complementary and non-substitutable.
+
+## Reference files
+
+- example: [`examples/ci-operational-checkpoint-v0.1.example.json`](examples/ci-operational-checkpoint-v0.1.example.json)
+- evaluator: [`conformance/ci_operational_checkpoint_reference.py`](conformance/ci_operational_checkpoint_reference.py)
+- tests: [`conformance/test_ci_operational_checkpoint.py`](conformance/test_ci_operational_checkpoint.py)
+- codes: [`CI-CHECKPOINT-CODES.md`](CI-CHECKPOINT-CODES.md)
+
+## Non-claims
+
+This profile is not durable storage, a distributed lease, proof of execution, or
+permission for a tool call.

standards/agent-continuity/README.md

@@ -1,6 +1,6 @@
 # Agent Continuity & Authority
 
-Vendor-neutral specifications and executable conformance checks for preserving operational continuity across context compaction, session restart, and cross-session handoff.
+Vendor-neutral specifications and executable conformance checks for preserving operational continuity across context compaction, session restart, cross-session handoff, and CI/CD phase transitions.
 
 ## RFC v0.1
 
@@ -10,6 +10,30 @@ defines a bounded, structured envelope that carries the active operational tail
 
 A separate restore-results document records which required reads and evidence checks were actually completed. The envelope declares the gate; restore results satisfy it.
 
+## CI Operational Checkpoint v0.1
+
+[`CI-OPERATIONAL-CHECKPOINT-V0.1.md`](./CI-OPERATIONAL-CHECKPOINT-V0.1.md)
+
+specializes the continuity boundary for agentic CI/CD work. It records exact
+workspace identity, goal, constraints, rejected approaches, verification state,
+and the next action while fixing authority to `context_only`.
+
+The profile deterministically returns:
+
+```text
+CONTINUE
+REVALIDATE_WORKSPACE
+RESTART_REQUIRED
+IDEMPOTENT_REPLAY
+REJECT_LINEAGE_MISMATCH
+REJECT_UNVERIFIED_COMPLETION
+REJECT_INVALID_AUTHORITY
+```
+
+A checkpoint may preserve the intent to merge or deploy, but it cannot authorize
+that action. A fresh Action Envelope and current exact-target evidence remain
+mandatory.
+
 ## Problem
 
 Coding agents may lose task continuity after compaction or handoff. They can repeat completed work, violate recent constraints, forget rejected approaches, or confidently reconstruct an execution history that is not supported by durable evidence.
@@ -21,10 +45,12 @@ Coding agents may lose task continuity after compaction or handoff. They can rep
 ## Package
 
 - RFC specification;
-- JSON Schemas for the envelope and restore results;
-- example envelope and restore results;
-- reference validator;
-- executable conformance tests.
+- CI operational checkpoint profile;
+- JSON Schemas for envelopes, restore results, and CI checkpoints;
+- complete examples;
+- deterministic reference validators;
+- executable conformance tests;
+- stable CI checkpoint outcome registry.
 
 ## Quick validation
 
@@ -39,16 +65,19 @@ python -m unittest discover \
 ## What the suite verifies
 
 - published JSON Schema enforcement;
-- canonical envelope digest integrity;
+- canonical envelope and checkpoint digest integrity;
 - trusted-source authority boundaries;
 - independent digest/receipt evidence checks;
 - required-read completion;
 - fail-closed restore behavior;
-- unresolved task verification remaining unresolved.
+- unresolved task verification remaining unresolved;
+- CI workspace drift detection;
+- lineage, replay, rejected-approach, and verification continuity;
+- fresh authorization requirements for merge and deploy intent.
 
 ## Intended integrations
 
-The specification is implementation-neutral. Codex, Claude Code, IDE agents, CLI agents, and multi-agent runtimes may store or transport the envelope differently while preserving the same observable guarantees.
+The specifications are implementation-neutral. Codex, Claude Code, IDE agents, CLI agents, CI bots, and multi-agent runtimes may store or transport the envelopes differently while preserving the same observable guarantees.
 
 ## Status
 

standards/agent-continuity/conformance/ci_operational_checkpoint_reference.py

@@ -0,0 +1,25 @@
+from __future__ import annotations
+
+from typing import Any, Iterable, Mapping
+
+import ci_operational_checkpoint_reference_core as _core
+from ci_operational_checkpoint_reference_core import *  # noqa: F401,F403
+
+
+def evaluate_resume(
+    checkpoint: Mapping[str, Any],
+    *,
+    current_workspace: Mapping[str, Any],
+    previous_checkpoint: Mapping[str, Any] | None = None,
+    seen_checkpoint_ids: Iterable[str] = (),
+    known_parent_ids: Iterable[str] = (),
+) -> dict[str, str]:
+    """Delegate to the canonical core evaluator without pre-validation."""
+
+    return _core.evaluate_resume(
+        checkpoint,
+        current_workspace=current_workspace,
+        previous_checkpoint=previous_checkpoint,
+        seen_checkpoint_ids=seen_checkpoint_ids,
+        known_parent_ids=known_parent_ids,
+    )

standards/agent-continuity/conformance/ci_operational_checkpoint_reference_core.py

@@ -0,0 +1,111 @@
+from __future__ import annotations
+
+from typing import Any, Iterable, Mapping
+
+import ci_operational_checkpoint_reference_impl as _impl
+from ci_operational_checkpoint_reference_impl import *  # noqa: F401,F403
+
+_result = _impl._result
+_schema_errors = _impl._schema_errors
+_ORIGINAL_PREVIOUS_CHECKPOINT_INTEGRITY_ERROR = (
+    _impl._previous_checkpoint_integrity_error
+)
+
+
+def _previous_checkpoint_integrity_error(
+    previous_checkpoint: Mapping[str, Any],
+) -> tuple[str, str] | None:
+    """Extend parent integrity after canonical schema and digest validation."""
+
+    error = _ORIGINAL_PREVIOUS_CHECKPOINT_INTEGRITY_ERROR(previous_checkpoint)
+    if error is not None:
+        return error
+
+    sequence = previous_checkpoint["sequence"]
+    parent_id = previous_checkpoint["parent_checkpoint_id"]
+    checkpoint_id = previous_checkpoint["checkpoint_id"]
+    if sequence == 0 and parent_id is not None:
+        return (
+            "PREVIOUS_CHECKPOINT_SEMANTIC_INVALID",
+            "root previous checkpoint must not declare a parent",
+        )
+    if sequence > 0 and parent_id is None:
+        return (
+            "PREVIOUS_CHECKPOINT_SEMANTIC_INVALID",
+            "non-root previous checkpoint must declare a parent",
+        )
+    if parent_id is not None and checkpoint_id == parent_id:
+        return (
+            "PREVIOUS_CHECKPOINT_SEMANTIC_INVALID",
+            "previous checkpoint cannot be its own parent",
+        )
+
+    for row in previous_checkpoint["verification"]["completed"]:
+        if not row["evidence_refs"]:
+            return (
+                "PREVIOUS_CHECKPOINT_SEMANTIC_INVALID",
+                f"{row['verification_id']} has no evidence references",
+            )
+
+    next_action = previous_checkpoint["next_action"]
+    if (
+        next_action["action_class"] in {"merge", "deploy"}
+        and not next_action["requires_fresh_authority"]
+    ):
+        return (
+            "PREVIOUS_CHECKPOINT_SEMANTIC_INVALID",
+            "parent merge or deploy intent requires fresh authority",
+        )
+    return None
+
+
+# Install the enhanced parent hook into the canonical ordered implementation.
+_impl._previous_checkpoint_integrity_error = _previous_checkpoint_integrity_error
+
+
+def evaluate_resume(
+    checkpoint: Mapping[str, Any],
+    *,
+    current_workspace: Mapping[str, Any],
+    previous_checkpoint: Mapping[str, Any] | None = None,
+    seen_checkpoint_ids: Iterable[str] = (),
+    known_parent_ids: Iterable[str] = (),
+) -> dict[str, str]:
+    """Preserve canonical validation order and stable authority diagnostics."""
+
+    errors = _schema_errors(checkpoint)
+    authority_error = next(
+        (
+            error
+            for error in errors
+            if list(error.absolute_path) == ["authority"]
+        ),
+        None,
+    )
+    if authority_error is not None:
+        return _result(
+            REJECT_INVALID_AUTHORITY,
+            "AUTHORITY_NOT_CONTEXT_ONLY",
+            authority_error.message,
+        )
+
+    result = _impl.evaluate_resume(
+        checkpoint,
+        current_workspace=current_workspace,
+        previous_checkpoint=previous_checkpoint,
+        seen_checkpoint_ids=seen_checkpoint_ids,
+        known_parent_ids=known_parent_ids,
+    )
+
+    if result["outcome"] == CONTINUE:
+        sequence = checkpoint["sequence"]
+        if (
+            sequence > 0
+            and checkpoint["checkpoint_id"] == checkpoint["parent_checkpoint_id"]
+        ):
+            return _result(
+                REJECT_LINEAGE_MISMATCH,
+                "CHECKPOINT_ID_REUSED",
+                "a checkpoint cannot reuse its parent checkpoint ID",
+            )
+    return result