Execution Log: instances scoped to their own work items

Append-only log of progress for the user's visibility.

Phase transitions

Phase	Entered	Reviewed/approved by	Notes
phase-selection	2026-09-08	—	Tier 4 (human-approves-pr, and a named human security sign-off at humanSignOffMinTier: 4): a seam on the dispatch path — the one place an event becomes a session — plus an additive block in cli-config.schema.json (an autonomy.sensitivePaths entry). Brainstorming skipped: the ticket's bullets are the requirements and its three questions are answered in requirements.md § Open questions. No authorized the-loop execute reaches this cloud session, so the selection is recorded here and every phase is walked
requirements-definition	2026-09-08		requirements.md — five requirements, six abuse cases
design	2026-09-08		design.md — one block, one seam, one token, one route; decision-110
test-planning	2026-09-08		testing-plan.md — thirteen rows, seven applicable
tasks-breakdown	2026-09-08		tasks.md — seven tasks
implementation	2026-09-08		On claude/github-issue-322-2yely4
verification	2026-09-08		evidence/verification.md — rows T1, T2, T3, T8, T10, T12; evidence/security-review.md — six abuse cases, six closed
needs-review	2026-09-08		PR raised; awaiting the owner (tier 4: human-approves-pr + the named security sign-off)
complete

Pull requests

PR	Scope / tasks	Status
#324	tasks 1–7: the whole work item	open

Progress entries

2026-09-08 — implemented, verified, ready for review

• Phase: implementation → verification → needs-review
• Did: tasks 1–7, red first each. the_loop.instance (InstanceConfig, parse_address, decide, the outcome constants); cli_config.apply_instance (the _ghBinary-style fan-out, applied only when a block is declared) and RoutingConfig.instance; the seam in Dispatcher.handle with _manages and _refuse_scope, the four outcomes added to SETTLED_OUTCOMES and to the poll.comment_settled catalogue entry; ControlRecord.instance on every record the dispatcher and core.sessions write; command_comment(address=) on the CLI's posted keyword, announcement_body(instance=), TmuxRunner.instance with the one-time tmux -V probe and the -e THE_LOOP_INSTANCE=<name> argv; core.sessions refusing a start on a locked instance before anything is recorded or posted; core/instance.py::describe_instance behind GET /api/v1/instance, the get_instance MCP tool, loop.instance() and the status line; the schema block (authored + packaged, byte-identical, purely additive), the template and this repository's config; the docs.
• Checkpoint/tests: make check — see evidence/verification.md. New tests: 60 unit (test_instance.py), 14 two-dispatcher scenarios (test_instance_integration.py), 1 status-line test. Existing assertions changed: one — the settled-vocabulary test in test_routing.py, which by its own docstring is where a new settlement is added.
• Self-review: three passes over the diff. Pass one found the fan-out stamping _instance onto every routing dict, which broke three equality tests on configs with no block — made conditional. Pass two verified three design claims against the code: only comment and review bodies reach parse_address (event_body returns None for an issue's own body, so an issue body cannot address); the receiver's and the poller's reload paths build from load_cli_config, so the block is hot-reloaded; the dashboard's config writer reads raw YAML, so the private key never reaches the file. Pass three found nothing new.
• Next: the owner's review and the tier-4 security sign-off.
• Blockers: none.

2026-09-08 — spec chain drafted

• Phase: requirements-definition → tasks-breakdown
• Did: read webhook/dispatcher.py (handle, _on_unmatched, _apply_control, _spawn_for, reload), control.py, cli_config.py, state.py, runner.py, announce.py, core/sessions.py, core/lifecycle.py, api/routes.py and the two daemons at 10a55b3; found that every instance judges every labelled event identically and that the one seam where an event becomes a session is handle; found the _ghBinary fan-out precedent for a top-level block reaching RoutingConfig; found the standing-session name grammar and the collaborator @login argument as the precedents for a validated token. Wrote the four artifacts and the decision.
• Checkpoint/tests: baseline — test_control.py green (57 passed).
• Next: task 1 (the block), red first.
• Blockers: none.

Verification results

Only when this work item declared test-planning away. It did not: results live in testing-plan.md.

What was verified	Command	Outcome	Evidence
—	—	—	see testing-plan.md

Design critic review

Not selected for this work item.

Round	Critic (<harness>/<model>)	Outcome	Findings → disposition	Link

Review cycles

Cycle	Type (self/critic/security)	Reviewer	Outcome	Link
1	self	the-loop (this session)	new findings — the unconditional _instance fan-out broke three equality tests on block-less configs: made conditional; the settled vocabulary needed the four outcomes in its test and catalogue entry: added	this log
2	self	the-loop (this session)	zero new findings — three design claims verified against the code (issue bodies never parsed; both reload paths carry the block; the config writer never sees the private key)	this log
3	self	the-loop (this session)	zero (converged)	this log
—	critic	—	unavailable — reviews.critics is empty in this repository's config; does not count toward criticReviewCount	—
4	security	the-loop checklist	pass on the autonomous review; human sign-off pending (tier 4)	evidence/security-review.md

Security review (gate)

• Mechanism: the-loop checklist (security.review.mechanism: auto; no security-review skill is invocable from this session's plugin set)
• Outcome: pass — evidence/security-review.md, six abuse cases closed
• Human sign-off: required and pending (tier 4 ≥ humanSignOffMinTier: 4) — the owner's approval of the PR is the named sign-off; requested in the PR briefing

Final validation evidence

Requirement	Proof
R1.1 the block's shape	test_the_block_reads_name_mode_and_declared_refs; make validate against the authored schema
R1.2 unset = 13.3.1, additive	test_the_block_unset_is_an_unnamed_open_instance, test_a_bare_routing_mapping_is_an_unnamed_open_instance, Scenario: An open, unnamed instance is 13.3.1; test_control_integration.py + test_graphlink_integration.py unchanged and green; CURRENT_CONFIG_VERSION unchanged
R1.3 in cli-config.yaml, hot-reloaded	test_the_block_reaches_routing_config_through_the_loaded_config; both daemons' reload paths build from load_cli_config (verified in self-review pass two)
R1.4, R1.5 narrowing	test_the_block_refuses_a_name_outside_the_grammar, test_an_unknown_mode_resolves_to_locked, test_the_block_addressed_without_a_name_is_locked, test_the_block_skips_a_bad_declaration_by_index_and_keeps_the_rest
R2.1 managed = as before, every mode	Scenario: A managed work item is in scope whatever the mode (×3), Scenario: A stopped work item is still this instance's; decide rows 3
R2.2 open = 13.3.1	decide row 4; the open-unnamed scenario
R2.3 addressed	Scenario: An addressed start reaches only the instance it names, Scenario: An unaddressed start on two addressed instances reaches neither
R2.4 locked	Scenario: A locked instance takes only what its config declares; test_an_address_does_not_unlock_a_locked_instance
R2.5 command → control.rejected	the addressed and unaddressed scenarios assert control.rejected with the reason and delivery_outcome == control-rejected
R2.6 no mark	Scenario: A non-owner leaves nothing behind (no portable/, no session, no spawn, no delivery, one dispatch.dropped)
R2.7 sessions start	Scenario: sessions start on a locked instance, Scenario: sessions start on an addressed instance
R3.1, R3.4 the token	test_the_token_is_read_as_a_whole_word_with_a_case_insensitive_prefix (six bodies), test_a_malformed_token_is_not_an_address_and_reaches_no_record (eleven)
R3.2 address is authoritative	Scenario: An explicit address wins over the managed set; decide row 2
R3.3 ambiguity	test_two_different_addresses_are_ambiguous, Scenario: A comment naming two instances is ambiguous
R3.5 both ingresses	the seam reads event_body, the one body accessor both ingresses feed (parse_control_command reads the same); the poller's RoutedEvent shape is what the scenarios use (labeled=False)
R4.1 control.instance	test_the_control_record_carries_the_instance_and_reads_old_records_as_unnamed; every scenario that spawns asserts the record names the instance
R4.2 the posted keyword	test_the_posted_keyword_carries_the_address; the addressed sessions start scenario
R4.3 the announcement	test_the_announcement_names_the_instance
R4.4 the route	test_describe_instance_merges_declared_session_and_control_sources, test_describe_instance_for_an_unset_block_is_unnamed_open_and_empty; test_api_contract_parity.py; test_mcp_integration.py; test_sdk_docs_parity.py
R4.5 status	test_status_prints_the_instance_line
R4.6 THE_LOOP_INSTANCE	test_a_named_instance_spawns_with_the_environment_variable, test_an_unnamed_instance_spawns_with_the_argv_it_used_before, test_an_old_tmux_gets_no_environment_flag_and_one_warning
R5.1–R5.3 docs	test_docs_parity.py (P3, P4, P5 over instance-options.md); make validate (template + this repo's config); markdownlint over 977 files
A1–A6	evidence/security-review.md

Capability docs

Capability doc	What changed	History row
instances.md	minted — the block, the managed set, the modes, the address token, the refusal, the record, the surface, what it does not do, and the seams left for the manager and the ticket-managed instance	issue-322 row
capabilities.md	the index row for instances	—
webhook-triggers.md	a current-behaviour bullet: the scope decision runs before anything else acts on an event, and can only refuse	issue-322 row
cli.md	a current-behaviour bullet: the status line, the locked refusal of sessions start, the address on posted keywords	issue-322 row
control-plane.md	a current-behaviour bullet: GET /api/v1/instance, get_instance, loop.instance() — the manager seam	issue-322 row

Documentation

Document	What changed
docs/config/cli/instance-options.md	new — name, scope.mode, scope.workItems with type, default, grammar and the three rules (parity-tested)
docs/config/cli/index.md	the options-by-area row
docs/cli/instances.md	new guide — the shape, the three modes, addressing, pinning and locking, what the record says, what the design does not do
docs/cli/index.md	a "you want to…" row pointing at the guide
docs/cli/state.md	control.instance in the field table and the example record
docs/cli/commands/status.md	the instance line, first in the example, and its JSON form
docs/.vitepress/config.mts	sidebar entries for the guide and the options page
docs/api-specs/openapi/the-loop.v1.yaml	GET /api/v1/instance (getInstance)
docs/sdk/reference.md	the instance() row
skills/the-loop/reference/automation.md	a bullet beside the control keywords: the block, the modes, the address token
skills/the-loop/templates/cli-config.yaml, .the-loop/cli-config.yaml	the instance block at its defaults, commented
.the-loop/cli-config.schema.json, cli/the_loop/schemas/cli-config.schema.json	the instance block (byte-identical copies; purely additive)
docs/decisions/decision-110.md, decisions.md	the decision and its index row
README.md, skills/the-loop/SKILL.md	unchanged — neither enumerates the daemon's configuration blocks; the operating model itself did not change (an instance is still armed by a label and started by an authorized keyword)