emrg: tool intent parameter replaces static purpose (rants 10:33:29 + 10:35:24)

[argszero]

Tool intent parameter replaces static purpose (host rants 10:33:29 + 10:35:24)

Host-finalized: every tool call should carry the agent's per-call intent ("why am I calling this, what do I want to achieve") instead of the static purpose field. Logs/GUI show the intent — real "doing X because of Y" readability.

Changes

1. 7 tools (bash/edit/glob/grep/read/write/submit_rant) — added required intent string param to their JSON Schemas (properties + required); removed the static purpose=... lines.
2. ToolDefinition — purpose field + docstring removed (tool_types.py).
3. Execution ignores intent — intent is display-only metadata; the executors keep running on their real arguments (unknown keys are naturally ignored by arguments.get(...)).
4. Daemon logging — 3 sites now log the intent:
   • main tool loop: tool call: <name> — <intent> (was — <purpose> (<redacted JSON>))
   • memory reflection + consolidation debug logs: same switch
   • intent missing (internal/LLM-free calls) → -
5. tool_start broadcast — carries "intent": args.get("intent"); ToolStart protocol gains an intent field.
6. TUI/GUI display — TUI ToolCard header shows intent when present (falls back to formatted args); GUI tool row shows an italic tool-intent span.
7. Prompt guidance — system.j2 Tool Usage section now instructs every tool call to include intent.

Tests

• test_tool_types.py::test_no_purpose_field — ToolDefinition has no purpose
• test_submit_rant_tool.py — test_all_tools_have_purpose → test_all_tools_require_intent (all 7 tools require intent, no purpose); consent-contract test asserts intent required
• test_protocol.py — ToolStart carries intent (and defaults "")
• test_ws_e2e.py::test_tool_intent_logged_and_ignored_by_execution — daemon broadcasts intent in tool_start while the executor still runs on real args
• Agent.md test count 967 → 969 (branch base; +2 net tests)

Verification

• py_compile all touched Python ✓ · node --check chat.js ✓
• Mocked run: tool call: bash — 检查回显 / missing intent → tool call: bash — - ✓
• All 7 tool schemas: intent in properties + required, no purpose attr ✓

[argszero]

✅ LGTM — cycle (1/3): head b1375e0, CI test + test-windows PASS (32211538781). Tool intent required param on all 7 tools (purpose removed), ToolStart.intent broadcast, daemon tool-call logging shows intent, TUI/GUI display intent; executors ignore the extra key. Branch merged master (v0.2.52 + #863) and Agent.md count synced to 971.

[argszero]

✅ LGTM — cycle (2/3): re-reviewed head b1375e0 fresh. All 7 tool executors read args via .get() (extra intent key tolerated, never executed); daemon logs/broadcasts intent (tool call + memory reflection + consolidation); ToolStart carries intent; TUI card header + GUI tool-row show intent; system.j2 instructs agent to always write intent; purpose field fully removed from ToolDefinition with updated tests (protocol defaults, no-purpose guard, e2e stateful two-round mock verifies broadcast + executor). CI test + test-windows PASS (32211538781).

[pm25coder]

Independent test report (cycle 2026-08-19, Contributor):

[argszero]

✅ LGTM — cycle (3/3): head b1375e0 unchanged since 2/3, CI test + test-windows PASS (32211538781), MERGEABLE/CLEAN. Fresh verification of the diff (21 files, +183/−41): intent required param on all 7 tools, purpose field removed, executors use .get() (intent never executed), daemon logs/broadcasts intent, TUI/GUI display, system.j2 instructs agent, e2e stateful mock asserts broadcast + execution. Approving for merge.

[argszero]

Tool intent parameter replaces static purpose (host rants 10:33:29 + 10:35:24)

Host-finalized: every tool call should carry the agent's per-call intent ("why am I calling this, what do I want to achieve") instead of the static purpose field. Logs/GUI show the intent — real "doing X because of Y" readability.

Changes

1. 7 tools (bash/edit/glob/grep/read/write/submit_rant) — added required intent string param to their JSON Schemas (properties + required); removed the static purpose=... lines.
2. ToolDefinition — purpose field + docstring removed (tool_types.py).
3. Execution ignores intent — intent is display-only metadata; the executors keep running on their real arguments (unknown keys are naturally ignored by arguments.get(...)).
4. Daemon logging — 3 sites now log the intent:
   • main tool loop: tool call: <name> — <intent> (was — <purpose> (<redacted JSON>))
   • memory reflection + consolidation debug logs: same switch
   • intent missing (internal/LLM-free calls) → -
5. tool_start broadcast — carries "intent": args.get("intent"); ToolStart protocol gains an intent field.
6. TUI/GUI display — TUI ToolCard header shows intent when present (falls back to formatted args); GUI tool row shows an italic tool-intent span.
7. Prompt guidance — system.j2 Tool Usage section now instructs every tool call to include intent.

Tests

• test_tool_types.py::test_no_purpose_field — ToolDefinition has no purpose
• test_submit_rant_tool.py — test_all_tools_have_purpose → test_all_tools_require_intent (all 7 tools require intent, no purpose); consent-contract test asserts intent required
• test_protocol.py — ToolStart carries intent (and defaults "")
• test_ws_e2e.py::test_tool_intent_logged_and_ignored_by_execution — daemon broadcasts intent in tool_start while the executor still runs on real args
• Agent.md test count 967 → 969 (branch base; +2 net tests)

Verification

• py_compile all touched Python ✓ · node --check chat.js ✓
• Mocked run: tool call: bash — 检查回显 / missing intent → tool call: bash — - ✓
• All 7 tool schemas: intent in properties + required, no purpose attr ✓

[argszero]

✅ LGTM — cycle (1/3): head b1375e0, CI test + test-windows PASS (32211538781). Tool intent required param on all 7 tools (purpose removed), ToolStart.intent broadcast, daemon tool-call logging shows intent, TUI/GUI display intent; executors ignore the extra key. Branch merged master (v0.2.52 + #863) and Agent.md count synced to 971.

[argszero]

✅ LGTM — cycle (2/3): re-reviewed head b1375e0 fresh. All 7 tool executors read args via .get() (extra intent key tolerated, never executed); daemon logs/broadcasts intent (tool call + memory reflection + consolidation); ToolStart carries intent; TUI card header + GUI tool-row show intent; system.j2 instructs agent to always write intent; purpose field fully removed from ToolDefinition with updated tests (protocol defaults, no-purpose guard, e2e stateful two-round mock verifies broadcast + executor). CI test + test-windows PASS (32211538781).

[pm25coder]

Independent test report (cycle 2026-08-19, Contributor):

[argszero]

✅ LGTM — cycle (3/3): head b1375e0 unchanged since 2/3, CI test + test-windows PASS (32211538781), MERGEABLE/CLEAN. Fresh verification of the diff (21 files, +183/−41): intent required param on all 7 tools, purpose field removed, executors use .get() (intent never executed), daemon logs/broadcasts intent, TUI/GUI display, system.j2 instructs agent, e2e stateful mock asserts broadcast + execution. Approving for merge.

[argszero]

Tool intent parameter replaces static purpose (host rants 10:33:29 + 10:35:24)

Host-finalized: every tool call should carry the agent's per-call intent ("why am I calling this, what do I want to achieve") instead of the static purpose field. Logs/GUI show the intent — real "doing X because of Y" readability.

Changes

1. 7 tools (bash/edit/glob/grep/read/write/submit_rant) — added required intent string param to their JSON Schemas (properties + required); removed the static purpose=... lines.
2. ToolDefinition — purpose field + docstring removed (tool_types.py).
3. Execution ignores intent — intent is display-only metadata; the executors keep running on their real arguments (unknown keys are naturally ignored by arguments.get(...)).
4. Daemon logging — 3 sites now log the intent:
   • main tool loop: tool call: <name> — <intent> (was — <purpose> (<redacted JSON>))
   • memory reflection + consolidation debug logs: same switch
   • intent missing (internal/LLM-free calls) → -
5. tool_start broadcast — carries "intent": args.get("intent"); ToolStart protocol gains an intent field.
6. TUI/GUI display — TUI ToolCard header shows intent when present (falls back to formatted args); GUI tool row shows an italic tool-intent span.
7. Prompt guidance — system.j2 Tool Usage section now instructs every tool call to include intent.

Tests

• test_tool_types.py::test_no_purpose_field — ToolDefinition has no purpose
• test_submit_rant_tool.py — test_all_tools_have_purpose → test_all_tools_require_intent (all 7 tools require intent, no purpose); consent-contract test asserts intent required
• test_protocol.py — ToolStart carries intent (and defaults "")
• test_ws_e2e.py::test_tool_intent_logged_and_ignored_by_execution — daemon broadcasts intent in tool_start while the executor still runs on real args
• Agent.md test count 967 → 969 (branch base; +2 net tests)

Verification

• py_compile all touched Python ✓ · node --check chat.js ✓
• Mocked run: tool call: bash — 检查回显 / missing intent → tool call: bash — - ✓
• All 7 tool schemas: intent in properties + required, no purpose attr ✓

[argszero]

✅ LGTM — cycle (1/3): head b1375e0, CI test + test-windows PASS (32211538781). Tool intent required param on all 7 tools (purpose removed), ToolStart.intent broadcast, daemon tool-call logging shows intent, TUI/GUI display intent; executors ignore the extra key. Branch merged master (v0.2.52 + #863) and Agent.md count synced to 971.

[argszero]

✅ LGTM — cycle (2/3): re-reviewed head b1375e0 fresh. All 7 tool executors read args via .get() (extra intent key tolerated, never executed); daemon logs/broadcasts intent (tool call + memory reflection + consolidation); ToolStart carries intent; TUI card header + GUI tool-row show intent; system.j2 instructs agent to always write intent; purpose field fully removed from ToolDefinition with updated tests (protocol defaults, no-purpose guard, e2e stateful two-round mock verifies broadcast + executor). CI test + test-windows PASS (32211538781).

[pm25coder]

Independent test report (cycle 2026-08-19, Contributor):

[argszero]

✅ LGTM — cycle (3/3): head b1375e0 unchanged since 2/3, CI test + test-windows PASS (32211538781), MERGEABLE/CLEAN. Fresh verification of the diff (21 files, +183/−41): intent required param on all 7 tools, purpose field removed, executors use .get() (intent never executed), daemon logs/broadcasts intent, TUI/GUI display, system.j2 instructs agent, e2e stateful mock asserts broadcast + execution. Approving for merge.

[argszero]

Tool intent parameter replaces static purpose (host rants 10:33:29 + 10:35:24)

Host-finalized: every tool call should carry the agent's per-call intent ("why am I calling this, what do I want to achieve") instead of the static purpose field. Logs/GUI show the intent — real "doing X because of Y" readability.

Changes

1. 7 tools (bash/edit/glob/grep/read/write/submit_rant) — added required intent string param to their JSON Schemas (properties + required); removed the static purpose=... lines.
2. ToolDefinition — purpose field + docstring removed (tool_types.py).
3. Execution ignores intent — intent is display-only metadata; the executors keep running on their real arguments (unknown keys are naturally ignored by arguments.get(...)).
4. Daemon logging — 3 sites now log the intent:
   • main tool loop: tool call: <name> — <intent> (was — <purpose> (<redacted JSON>))
   • memory reflection + consolidation debug logs: same switch
   • intent missing (internal/LLM-free calls) → -
5. tool_start broadcast — carries "intent": args.get("intent"); ToolStart protocol gains an intent field.
6. TUI/GUI display — TUI ToolCard header shows intent when present (falls back to formatted args); GUI tool row shows an italic tool-intent span.
7. Prompt guidance — system.j2 Tool Usage section now instructs every tool call to include intent.

Tests

• test_tool_types.py::test_no_purpose_field — ToolDefinition has no purpose
• test_submit_rant_tool.py — test_all_tools_have_purpose → test_all_tools_require_intent (all 7 tools require intent, no purpose); consent-contract test asserts intent required
• test_protocol.py — ToolStart carries intent (and defaults "")
• test_ws_e2e.py::test_tool_intent_logged_and_ignored_by_execution — daemon broadcasts intent in tool_start while the executor still runs on real args
• Agent.md test count 967 → 969 (branch base; +2 net tests)

Verification

• py_compile all touched Python ✓ · node --check chat.js ✓
• Mocked run: tool call: bash — 检查回显 / missing intent → tool call: bash — - ✓
• All 7 tool schemas: intent in properties + required, no purpose attr ✓

[argszero]

✅ LGTM — cycle (1/3): head b1375e0, CI test + test-windows PASS (32211538781). Tool intent required param on all 7 tools (purpose removed), ToolStart.intent broadcast, daemon tool-call logging shows intent, TUI/GUI display intent; executors ignore the extra key. Branch merged master (v0.2.52 + #863) and Agent.md count synced to 971.

[argszero]

✅ LGTM — cycle (2/3): re-reviewed head b1375e0 fresh. All 7 tool executors read args via .get() (extra intent key tolerated, never executed); daemon logs/broadcasts intent (tool call + memory reflection + consolidation); ToolStart carries intent; TUI card header + GUI tool-row show intent; system.j2 instructs agent to always write intent; purpose field fully removed from ToolDefinition with updated tests (protocol defaults, no-purpose guard, e2e stateful two-round mock verifies broadcast + executor). CI test + test-windows PASS (32211538781).

[pm25coder]

Independent test report (cycle 2026-08-19, Contributor):

[argszero]

✅ LGTM — cycle (3/3): head b1375e0 unchanged since 2/3, CI test + test-windows PASS (32211538781), MERGEABLE/CLEAN. Fresh verification of the diff (21 files, +183/−41): intent required param on all 7 tools, purpose field removed, executors use .get() (intent never executed), daemon logs/broadcasts intent, TUI/GUI display, system.j2 instructs agent, e2e stateful mock asserts broadcast + execution. Approving for merge.

[argszero]

Tool intent parameter replaces static purpose (host rants 10:33:29 + 10:35:24)

Host-finalized: every tool call should carry the agent's per-call intent ("why am I calling this, what do I want to achieve") instead of the static purpose field. Logs/GUI show the intent — real "doing X because of Y" readability.

Changes

1. 7 tools (bash/edit/glob/grep/read/write/submit_rant) — added required intent string param to their JSON Schemas (properties + required); removed the static purpose=... lines.
2. ToolDefinition — purpose field + docstring removed (tool_types.py).
3. Execution ignores intent — intent is display-only metadata; the executors keep running on their real arguments (unknown keys are naturally ignored by arguments.get(...)).
4. Daemon logging — 3 sites now log the intent:
   • main tool loop: tool call: <name> — <intent> (was — <purpose> (<redacted JSON>))
   • memory reflection + consolidation debug logs: same switch
   • intent missing (internal/LLM-free calls) → -
5. tool_start broadcast — carries "intent": args.get("intent"); ToolStart protocol gains an intent field.
6. TUI/GUI display — TUI ToolCard header shows intent when present (falls back to formatted args); GUI tool row shows an italic tool-intent span.
7. Prompt guidance — system.j2 Tool Usage section now instructs every tool call to include intent.

Tests

• test_tool_types.py::test_no_purpose_field — ToolDefinition has no purpose
• test_submit_rant_tool.py — test_all_tools_have_purpose → test_all_tools_require_intent (all 7 tools require intent, no purpose); consent-contract test asserts intent required
• test_protocol.py — ToolStart carries intent (and defaults "")
• test_ws_e2e.py::test_tool_intent_logged_and_ignored_by_execution — daemon broadcasts intent in tool_start while the executor still runs on real args
• Agent.md test count 967 → 969 (branch base; +2 net tests)

Verification

• py_compile all touched Python ✓ · node --check chat.js ✓
• Mocked run: tool call: bash — 检查回显 / missing intent → tool call: bash — - ✓
• All 7 tool schemas: intent in properties + required, no purpose attr ✓

[argszero]

✅ LGTM — cycle (1/3): head b1375e0, CI test + test-windows PASS (32211538781). Tool intent required param on all 7 tools (purpose removed), ToolStart.intent broadcast, daemon tool-call logging shows intent, TUI/GUI display intent; executors ignore the extra key. Branch merged master (v0.2.52 + #863) and Agent.md count synced to 971.

[argszero]

✅ LGTM — cycle (2/3): re-reviewed head b1375e0 fresh. All 7 tool executors read args via .get() (extra intent key tolerated, never executed); daemon logs/broadcasts intent (tool call + memory reflection + consolidation); ToolStart carries intent; TUI card header + GUI tool-row show intent; system.j2 instructs agent to always write intent; purpose field fully removed from ToolDefinition with updated tests (protocol defaults, no-purpose guard, e2e stateful two-round mock verifies broadcast + executor). CI test + test-windows PASS (32211538781).

[pm25coder]

Independent test report (cycle 2026-08-19, Contributor):

[argszero]

✅ LGTM — cycle (3/3): head b1375e0 unchanged since 2/3, CI test + test-windows PASS (32211538781), MERGEABLE/CLEAN. Fresh verification of the diff (21 files, +183/−41): intent required param on all 7 tools, purpose field removed, executors use .get() (intent never executed), daemon logs/broadcasts intent, TUI/GUI display, system.j2 instructs agent, e2e stateful mock asserts broadcast + execution. Approving for merge.

[argszero]

Tool intent parameter replaces static purpose (host rants 10:33:29 + 10:35:24)

Host-finalized: every tool call should carry the agent's per-call intent ("why am I calling this, what do I want to achieve") instead of the static purpose field. Logs/GUI show the intent — real "doing X because of Y" readability.

Changes

1. 7 tools (bash/edit/glob/grep/read/write/submit_rant) — added required intent string param to their JSON Schemas (properties + required); removed the static purpose=... lines.
2. ToolDefinition — purpose field + docstring removed (tool_types.py).
3. Execution ignores intent — intent is display-only metadata; the executors keep running on their real arguments (unknown keys are naturally ignored by arguments.get(...)).
4. Daemon logging — 3 sites now log the intent:
   • main tool loop: tool call: <name> — <intent> (was — <purpose> (<redacted JSON>))
   • memory reflection + consolidation debug logs: same switch
   • intent missing (internal/LLM-free calls) → -
5. tool_start broadcast — carries "intent": args.get("intent"); ToolStart protocol gains an intent field.
6. TUI/GUI display — TUI ToolCard header shows intent when present (falls back to formatted args); GUI tool row shows an italic tool-intent span.
7. Prompt guidance — system.j2 Tool Usage section now instructs every tool call to include intent.

Tests

• test_tool_types.py::test_no_purpose_field — ToolDefinition has no purpose
• test_submit_rant_tool.py — test_all_tools_have_purpose → test_all_tools_require_intent (all 7 tools require intent, no purpose); consent-contract test asserts intent required
• test_protocol.py — ToolStart carries intent (and defaults "")
• test_ws_e2e.py::test_tool_intent_logged_and_ignored_by_execution — daemon broadcasts intent in tool_start while the executor still runs on real args
• Agent.md test count 967 → 969 (branch base; +2 net tests)

Verification

• py_compile all touched Python ✓ · node --check chat.js ✓
• Mocked run: tool call: bash — 检查回显 / missing intent → tool call: bash — - ✓
• All 7 tool schemas: intent in properties + required, no purpose attr ✓

[argszero]

✅ LGTM — cycle (1/3): head b1375e0, CI test + test-windows PASS (32211538781). Tool intent required param on all 7 tools (purpose removed), ToolStart.intent broadcast, daemon tool-call logging shows intent, TUI/GUI display intent; executors ignore the extra key. Branch merged master (v0.2.52 + #863) and Agent.md count synced to 971.

[argszero]

✅ LGTM — cycle (2/3): re-reviewed head b1375e0 fresh. All 7 tool executors read args via .get() (extra intent key tolerated, never executed); daemon logs/broadcasts intent (tool call + memory reflection + consolidation); ToolStart carries intent; TUI card header + GUI tool-row show intent; system.j2 instructs agent to always write intent; purpose field fully removed from ToolDefinition with updated tests (protocol defaults, no-purpose guard, e2e stateful two-round mock verifies broadcast + executor). CI test + test-windows PASS (32211538781).

[pm25coder]

Independent test report (cycle 2026-08-19, Contributor):

[argszero]

✅ LGTM — cycle (3/3): head b1375e0 unchanged since 2/3, CI test + test-windows PASS (32211538781), MERGEABLE/CLEAN. Fresh verification of the diff (21 files, +183/−41): intent required param on all 7 tools, purpose field removed, executors use .get() (intent never executed), daemon logs/broadcasts intent, TUI/GUI display, system.j2 instructs agent, e2e stateful mock asserts broadcast + execution. Approving for merge.

[argszero]

Tool intent parameter replaces static purpose (host rants 10:33:29 + 10:35:24)

Host-finalized: every tool call should carry the agent's per-call intent ("why am I calling this, what do I want to achieve") instead of the static purpose field. Logs/GUI show the intent — real "doing X because of Y" readability.

Changes

1. 7 tools (bash/edit/glob/grep/read/write/submit_rant) — added required intent string param to their JSON Schemas (properties + required); removed the static purpose=... lines.
2. ToolDefinition — purpose field + docstring removed (tool_types.py).
3. Execution ignores intent — intent is display-only metadata; the executors keep running on their real arguments (unknown keys are naturally ignored by arguments.get(...)).
4. Daemon logging — 3 sites now log the intent:
   • main tool loop: tool call: <name> — <intent> (was — <purpose> (<redacted JSON>))
   • memory reflection + consolidation debug logs: same switch
   • intent missing (internal/LLM-free calls) → -
5. tool_start broadcast — carries "intent": args.get("intent"); ToolStart protocol gains an intent field.
6. TUI/GUI display — TUI ToolCard header shows intent when present (falls back to formatted args); GUI tool row shows an italic tool-intent span.
7. Prompt guidance — system.j2 Tool Usage section now instructs every tool call to include intent.

Tests

• test_tool_types.py::test_no_purpose_field — ToolDefinition has no purpose
• test_submit_rant_tool.py — test_all_tools_have_purpose → test_all_tools_require_intent (all 7 tools require intent, no purpose); consent-contract test asserts intent required
• test_protocol.py — ToolStart carries intent (and defaults "")
• test_ws_e2e.py::test_tool_intent_logged_and_ignored_by_execution — daemon broadcasts intent in tool_start while the executor still runs on real args
• Agent.md test count 967 → 969 (branch base; +2 net tests)

Verification

• py_compile all touched Python ✓ · node --check chat.js ✓
• Mocked run: tool call: bash — 检查回显 / missing intent → tool call: bash — - ✓
• All 7 tool schemas: intent in properties + required, no purpose attr ✓

[argszero]

✅ LGTM — cycle (1/3): head b1375e0, CI test + test-windows PASS (32211538781). Tool intent required param on all 7 tools (purpose removed), ToolStart.intent broadcast, daemon tool-call logging shows intent, TUI/GUI display intent; executors ignore the extra key. Branch merged master (v0.2.52 + #863) and Agent.md count synced to 971.

[argszero]

✅ LGTM — cycle (2/3): re-reviewed head b1375e0 fresh. All 7 tool executors read args via .get() (extra intent key tolerated, never executed); daemon logs/broadcasts intent (tool call + memory reflection + consolidation); ToolStart carries intent; TUI card header + GUI tool-row show intent; system.j2 instructs agent to always write intent; purpose field fully removed from ToolDefinition with updated tests (protocol defaults, no-purpose guard, e2e stateful two-round mock verifies broadcast + executor). CI test + test-windows PASS (32211538781).

[pm25coder]

Independent test report (cycle 2026-08-19, Contributor):

[argszero]

✅ LGTM — cycle (3/3): head b1375e0 unchanged since 2/3, CI test + test-windows PASS (32211538781), MERGEABLE/CLEAN. Fresh verification of the diff (21 files, +183/−41): intent required param on all 7 tools, purpose field removed, executors use .get() (intent never executed), daemon logs/broadcasts intent, TUI/GUI display, system.j2 instructs agent, e2e stateful mock asserts broadcast + execution. Approving for merge.

[argszero]

Tool intent parameter replaces static purpose (host rants 10:33:29 + 10:35:24)

Host-finalized: every tool call should carry the agent's per-call intent ("why am I calling this, what do I want to achieve") instead of the static purpose field. Logs/GUI show the intent — real "doing X because of Y" readability.

Changes

1. 7 tools (bash/edit/glob/grep/read/write/submit_rant) — added required intent string param to their JSON Schemas (properties + required); removed the static purpose=... lines.
2. ToolDefinition — purpose field + docstring removed (tool_types.py).
3. Execution ignores intent — intent is display-only metadata; the executors keep running on their real arguments (unknown keys are naturally ignored by arguments.get(...)).
4. Daemon logging — 3 sites now log the intent:
   • main tool loop: tool call: <name> — <intent> (was — <purpose> (<redacted JSON>))
   • memory reflection + consolidation debug logs: same switch
   • intent missing (internal/LLM-free calls) → -
5. tool_start broadcast — carries "intent": args.get("intent"); ToolStart protocol gains an intent field.
6. TUI/GUI display — TUI ToolCard header shows intent when present (falls back to formatted args); GUI tool row shows an italic tool-intent span.
7. Prompt guidance — system.j2 Tool Usage section now instructs every tool call to include intent.

Tests

• test_tool_types.py::test_no_purpose_field — ToolDefinition has no purpose
• test_submit_rant_tool.py — test_all_tools_have_purpose → test_all_tools_require_intent (all 7 tools require intent, no purpose); consent-contract test asserts intent required
• test_protocol.py — ToolStart carries intent (and defaults "")
• test_ws_e2e.py::test_tool_intent_logged_and_ignored_by_execution — daemon broadcasts intent in tool_start while the executor still runs on real args
• Agent.md test count 967 → 969 (branch base; +2 net tests)

Verification

• py_compile all touched Python ✓ · node --check chat.js ✓
• Mocked run: tool call: bash — 检查回显 / missing intent → tool call: bash — - ✓
• All 7 tool schemas: intent in properties + required, no purpose attr ✓

[argszero]

✅ LGTM — cycle (1/3): head b1375e0, CI test + test-windows PASS (32211538781). Tool intent required param on all 7 tools (purpose removed), ToolStart.intent broadcast, daemon tool-call logging shows intent, TUI/GUI display intent; executors ignore the extra key. Branch merged master (v0.2.52 + #863) and Agent.md count synced to 971.

[argszero]

✅ LGTM — cycle (2/3): re-reviewed head b1375e0 fresh. All 7 tool executors read args via .get() (extra intent key tolerated, never executed); daemon logs/broadcasts intent (tool call + memory reflection + consolidation); ToolStart carries intent; TUI card header + GUI tool-row show intent; system.j2 instructs agent to always write intent; purpose field fully removed from ToolDefinition with updated tests (protocol defaults, no-purpose guard, e2e stateful two-round mock verifies broadcast + executor). CI test + test-windows PASS (32211538781).

[pm25coder]

Independent test report (cycle 2026-08-19, Contributor):

[argszero]

✅ LGTM — cycle (3/3): head b1375e0 unchanged since 2/3, CI test + test-windows PASS (32211538781), MERGEABLE/CLEAN. Fresh verification of the diff (21 files, +183/−41): intent required param on all 7 tools, purpose field removed, executors use .get() (intent never executed), daemon logs/broadcasts intent, TUI/GUI display, system.j2 instructs agent, e2e stateful mock asserts broadcast + execution. Approving for merge.