Parfii-bot
|
e4b64418fc
|
feat(convergence/u2): capability renames + back-compat aliases
Pre-unlock wave U2. Task 3 from CONVERGENCE-PLAN — rename misleading
capability names, keep old names as deprecated aliases.
Renames:
- tools::read-only → tools::deny-tools (mechanism is tool-name denial,
not "read-only" metaphor)
- tools::cargo-only-bash → tools::bash-allowlist (mechanism is Bash
pattern allow-list; cargo-only is one config value)
Back-compat via registry.resolve_alias():
- Old dir _capabilities/tools/{read-only,cargo-only-bash}/ retained with
capability.toml-only stub: `alias = "<new-name>"` + `deprecated` field
- registry.rs loads alias stubs, redirects lookup before dispatch
- warn_deprecated_once() emits single-shot stderr per alias per process
via OnceLock<Mutex<HashSet>>
- Zero breaking change to existing manifests / task.toml referencing
old names
Rust impl files renamed in place:
- gates/tools_read_only.rs → gates/tools_deny_tools.rs (struct
DenyTools)
- gates/tools_cargo_only_bash.rs → gates/tools_bash_allowlist.rs
(struct BashAllowlist)
- gates/mod.rs + registry.rs + gate_smoke.rs updated
Roles updated (3): read-only.toml, explorer.toml, edit-local.toml —
reference new names directly.
Tests: kei-agent-runtime 41/41 (was 40, +1 deprecated_aliases_resolve
_to_new_names), _assembler 40/40 unchanged (substrate role expansion
follows new paths).
Docs updated: AGENT-ROLES.md, AGENT-SUBSTRATE-SCHEMA.md, 4 _manifests
referencing the old names (comment-only annotations).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
2026-04-23 03:43:40 +08:00 |
|
Parfii-bot
|
d72ae51f16
|
feat(agent-substrate/wrapper): kei-agent-runtime prepare — orchestrator ergonomics
Single-command "prepare spawn" that emits everything orchestrator needs
to invoke the Agent tool: composed prompt, subagent_type (from role's
new claude-subagent-type field), isolation mode, verify command,
ledger row.
Before this: orchestrator ran compose + read prompt + manually
constructed Agent tool call + manually built verify command. 4 steps.
After: `kei-agent-runtime prepare <task.toml> --format=human` outputs
a single copy-paste-ready block. Orchestrator pastes into Agent tool
and records the verify command for return.
Files:
- src/prepare.rs (170 LOC) — prepare() returns AgentInvocation struct
(agent_id, prompt, subagent_type, isolation, description,
verify_command, ledger_row)
- src/main.rs (+39 LOC) — Prepare subcommand with --format=human|json|toml
- src/lib.rs (+2 LOC — pub mod prepare)
- _roles/*.toml (5 files) — new optional claude-subagent-type field:
- edit-local / edit-shared → "code-implementer"
- read-only → "critic" (default; "architect" override possible)
- explorer → "Explore"
- git-ops → "NOT-SPAWNABLE" (refused by prepare with RULE 0.13)
- tests/prepare_smoke.rs (3 tests) — happy path, unknown role, non-spawnable refusal
- docs/AGENT-SUBSTRATE-SCHEMA.md (+ ## Orchestrator ergonomics section)
Tests: 40/40 (was 37, +3 prepare_smoke). Same path exercised in tempfile
fixtures that the real CLI would hit end-to-end.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
2026-04-23 03:25:14 +08:00 |
|
Parfii-bot
|
b82e3b039e
|
feat(agent-substrate/phase-3): kei-agent-runtime + kei-capability binaries
Two new crates implementing the substrate runtime per locked §Runtime
execution contract + §Capability trait contract (Rust) + §Verify
execution worktree→simulated-merge.
kei-agent-runtime — library + CLI binary:
- src/capability.rs — Capability trait (name/check/verify) + GateContext
+ GateDecision + VerifyContext + VerifyResult + RunMode + TaskSpec
- src/registry.rs — &str → &'static dyn Capability dispatch for 14 impls
- src/gates/ — 6 PreToolUse modules (policy::no-git-ops,
scope::files-{whitelist,denylist}, safety::no-dep-bump,
tools::read-only, tools::cargo-only-bash)
- src/verifies/ — 8 on-return modules (quality::constructor-pattern,
quality::cargo-check-green, quality::tests-green, safety::no-dep-bump,
scope::files-{whitelist,denylist}, output::{report-format,severity-grade})
- src/compose.rs — task.toml + role + capabilities → prompt.md
- src/spawn.rs — ledger fork + prompt write (actual Agent invocation
remains orchestrator's tool call)
- src/verify.rs — runs all capability verifies per role; collects
VerifyReport {passed, failed}
- src/simulated_merge.rs — git worktree add test-merge/<id> + apply diff
+ run verify; cleanup on Drop
- src/main.rs — clap CLI: compose | spawn | verify | run
kei-capability — thin CLI adapter crate:
- Depends on kei-agent-runtime path dep
- Subcommand `check <cap-name>` (PreToolUse gate; stdin JSON, exit 0|2)
- Subcommand `verify <cap-name>` (on-return; env-driven, exit 0 or fail)
- Pattern: shell hook = 3-line `exec kei-capability check "$CAP_NAME"`
Workspace Cargo.toml: both crates registered as members (under agent
substrate v1 marker).
cargo check --workspace: PASS
cargo test -p kei-agent-runtime: 37/37 green
- 6 capability_trait_smoke (registry lookups, unknown name → None)
- 3 compose_smoke (fixture role + caps → composed prompt)
- 12 gate_smoke (each gate: happy + deny + bypass)
- 4 simulated_merge_smoke (git worktree lifecycle)
- 12 verify_smoke (each verify: pass + fail + edge cases)
cargo test -p kei-capability: 0/0 (CLI binary, tested via lib)
(Agent completion report cut off by rate-limit at 60 tool-uses; code
itself is green — verified by orchestrator post-commit.)
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
2026-04-23 02:35:53 +08:00 |
|