CCR v1.5.0 - Mission Runtime P2 and first-time-agent docs

Collective Capability Runtime v1.5.0

CCR v1.5.0 completes the Mission Runtime P2 surface and makes the project easier for a first-time agent to use safely.

CCR is a local, JSON-first runtime for coordinating AI-agent work. It stores capability packets, tasks, residuals, verification evidence, mission reports, and workbench views. It does not create real ASI, grant execution authority, dispatch providers by default, or prove physical outcomes.

What is new

• Mission quickstart: ccr asi quickstart creates a local ASI-proxy mission fixture without network calls or provider execution.
• First-time-agent docs: README, docs index, getting-started guide, and command map now explain what to run first and what each command can write locally.
• Safer agent explanation: ccr agent explain --json now returns docs routes, first-use commands, P0/P1/P2 surfaces, safe boundaries, and mutation boundaries.
• P1 local gates: MCP descriptor/invocation, A2A card/handoff, external ingest, and provider manifest/conformance checks are local-first and non-dispatching.
• P2 usability surfaces: residual market/bounty/diff, static workbench export, operation replay/observation verification, CCR/PIC conformance checks, and provider registry validation.
• Release readiness: repo audit, release audit, CI, docs readiness tests, and package metadata now cover the v1.5.0 surface.

Safe first run

python -m pip install collective-capability-runtime
ccr agent explain --json
ccr asi quickstart --profile development --json
ccr mission next --mission mission:quickstart --compact --json
ccr workbench report --mission mission:quickstart --format markdown --out CCR_WORKBENCH.md
ccr residual market --mission mission:quickstart --json

Safety boundaries

• CCR does not detect or create real ASI.
• CCR does not grant execution authority.
• PIC/provider output is evidence only, not settlement.
• MCP/A2A descriptors or handoffs are evidence only, not delegated authority.
• Operation replay is not dispatch.
• Observation verification is not physical outcome proof.
• Static workbench is presentation, not proof.
• Residual market ranks work; it does not waive residuals.
• Provider registry is static metadata, not authority.
• Cache/index hits are not proof.

Validation before release

The release commit passed:

• ruff format --check, ruff check, mypy, compileall
• pytest with 150 tests
• schema validation smoke checks
• P0/P1/P2 runtime smoke checks
• ccr audit repo --json
• ccr audit release --dist dist --json
• twine check dist/*
• PIC compatibility audit in CI

CCR v1.4.0 - ASI-proxy loop runtime

v1.4.0 - ASI-proxy loop runtime

This release extends CCR as the runtime side of the bounded PIC/PIC-TS ASI-proxy loop. It imports candidate reports, schedules residual repair, maintains diagnostic reserve, and records runtime state without claiming settlement or real-world execution proof.

For first-time agents

• Start with docs/asi-proxy-loop.md, docs/agent-loop-protocol.md, and examples/asi_proxy_loop_bundle/.
• Use ccr loop init, ccr loop next, ccr loop import-report, ccr loop export-bundle, ccr token import, and ccr performance report to run the local loop.
• CCR tasks, active cuts, and safe commands are advisory. They do not dispatch providers, operate physical systems, or prove ASI.

Highlights

• Adds loop commands for init/status/next/step/import/export/bench/doctor.
• Adds token import/distillation/dedup, graph quotient, foundry active-cut smoothing, performance/cache reports, and SQLite public indices.
• Updates PIC compatibility audit for PIC v0.9.0 and keeps candidate-only residuals explicit.
• Adds schemas, docs, and example fixtures aligned with PIC and PIC-TS v0.9.0.
• Hardens CI and publish workflows with pinned actions, reduced permissions, and release artifact audits.

Verification

Local and GitHub CI passed for ruff, format, mypy, pytest, release audit, PIC compatibility audit, wheel/sdist build, twine check, pip-audit, bandit, and zizmor.

CCR v1.3.0 - ASI-proxy runtime controls

v1.3.0 - ASI-proxy runtime controls

This release makes CCR a portable runtime surface for PIC/CCR agents that need ASI-proxy acceleration diagnostics without converting diagnostics into uncontrolled execution.

Highlights:

• Adds fail-closed CARA/ASI-proxy phase acceleration checks for target validity, baseline observability, capital witness admission, and raw-net floors.
• Adds ccr foundry allocate --strategy phase-response --response-report <json> and ccr foundry simulate-allocation --cuts <json> --budget <json> for advisory phase-response allocation.
• Strengthens runtime capital witness normalization for negative liquidity, stale lifecycle evidence, and stale authority.
• Adds cross-repo fixtures and roundtrip coverage for PIC phase-response reports and CCR allocation simulation.

Safety boundary:
CCR reports and foundry allocations are advisory evidence. They do not authorize physical or external-world action by themselves. TRC authority, hazard envelope, rollback, lifecycle, logging, and explicit operation-gate checks remain mandatory before any real-world dispatch.

collective-capability-runtime v1.2.0

What changed

CCR v1.2.0 adds a TRC operation preflight path for agents consuming PIC trace-check or operation-gate reports.

New commands:

ccr operation preflight --trace pic_report.json --provider http --config provider.json --json
ccr operation dispatch --plan trc_operation_plan.json --provider http --config provider.json --execute --json
ccr operation observe --dispatch-report dispatch.json --observation observation.json --json

Agent-readable safety model

• CCR rechecks authority freshness and validity-domain scope before marking a plan ready.
• Provider dispatch requires a ready PIC/CCR plan, explicit provider config, allow_execute=true, a non-dry-run side-effect policy, and an operator approval reference.
• Fixture benchmark traces remain blocked and non-executable.
• Preflight is not dispatch. Dispatch is not physical outcome proof. Observation evidence still requires verifier acceptance before outcome claims.

Interop

CCR now accepts pic.trc_trace_report.v1 and pic.trc_operation_gate_report.v1, and publishes schemas for ccr.trc_operation_preflight.v1 and ccr.trc_operation_observation.v1.

Validation

Local and GitHub CI passed: format, lint, mypy, compileall, pytest, schema validation, phase-form smoke, repository audit, build, release audit, twine check, Bandit, and pip-audit.

v1.1.0

CCR v1.1.0 adds the runtime side of PIC/CCR ASI-proxy acceleration: agents can coordinate work, preserve residuals, and turn PIC trace-check reports into explicit dry-run operation plans.

What changed:

• Added workcell, distillation, residual ranking/repair, scheduler, foundry, and experiment helper commands.
• Added ccr operation plan and ccr operation execute for TRC-governed operation plans. Dispatch is dry-run unless a provider config and explicit --execute are supplied.
• Added ccr.trc_operation_plan.v1 schema validation and an ASI-proxy benchmark bundle.
• Updated PIC compatibility audit to accept the existing PIC v0.5 source tree during rollout and check PIC v0.6 interop markers when available.

Safety boundary:

• PIC reports, safe commands, and TRC operation plans are provider evidence and task inputs, not CCR settlement or automatic execution authority.
• settled=false remains expected while residuals or runtime gates remain open.

Install or update:
python -m pip install -U collective-capability-runtime

v1.0.0

Initial public release of Collective Capability Runtime (CCR).

• Agent-native Python runtime for task leasing, packet submission, verification orchestration, residual tracking, and ASI-proxy phase formation.
• SQLite + JSON artifact storage, deterministic schemas, effective graph observation, baseline comparison, and certificate-candidate generation.
• Optional PIC provider integration with kadubon/percolation-inversion-compiler; PIC output is imported as compatibility evidence and never settles CCR by itself.
• Release hardening includes CI, static checks, schema/example validation, PIC compatibility audit, public release audit, clean package build, and PyPI Trusted Publishing readiness.

Non-claims: CCR does not detect or create real ASI, update model weights, grant authority, bypass safety boundaries, or perform hidden external execution.