ayder-cli 2.4.0__tar.gz → 2.5.0__tar.gz

{ayder_cli-2.4.0 → ayder_cli-2.5.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: ayder-cli
-Version: 2.4.0
+Version: 2.5.0
 Summary: AI agent for any LLMs
 Project-URL: Homepage, https://github.com/ayder/ayder-cli
 Project-URL: Repository, https://github.com/ayder/ayder-cli.git
@@ -204,10 +204,11 @@ editor = "vim"                # Editor for /task-edit command
 verbose = false               # Show file contents after write + LLM debug
 max_background_processes = 5  # Max concurrent background processes (1-20)
 max_iterations = 10           # Max agentic iterations per message (1-100)
-max_output_tokens = 4096      # Max tokens in LLM response
+max_output_tokens = 32768     # Max tokens in LLM response
 max_history_messages = 30     # Messages kept in history
 prompt = "STANDARD"           # System prompt tier: MINIMAL, STANDARD, EXTENDED
 tool_tags = ["core", "metadata"]  # Enabled tool tags (see /plugin)
+# skills = ["engineering"]    # or "none"; default: the built-in skills
 agent_timeout = 300           # Seconds before a background agent is cancelled
 
 [logging]
@@ -353,7 +354,7 @@ ayder -f instructions.txt
 ayder --stdin < prompt.txt
 
 # Override the main LLM system prompt with a file
-# (tool instructions + project structure are still appended)
+# (tool instructions are still appended)
 ayder --system-prompt prompt-file.md "refactor this code"
 ```
 
@@ -414,12 +415,13 @@ ayder-cli includes a context manager that automatically keeps your conversation
 - **Real token counts**: Uses Ollama's `prompt_eval_count` and `eval_count` from native API responses instead of estimates. Compaction decisions are based on actual usage, not heuristics.
 - **Cache monitoring**: Tracks `prompt_eval_duration / prompt_eval_count` ratio to detect KV-cache hits (hot/warm/miss). When the cache is hot, compaction is delayed (threshold raised from 70% to 90%) to avoid invalidating it.
 - **Auto-detected context length**: On startup, queries `/api/show` to get the model's real context length instead of relying on `num_ctx`. Falls back to `num_ctx` if detection fails.
-- **Heuristic compaction**: When the context fills up, oldest message units are replaced with a text summary — no LLM call needed.
+- **Heuristic compaction**: When the context fills up, the oldest half of the message units is replaced with a text summary — no LLM call needed. The newest unit is never folded. The summary persists: every later request reuses the same summary message, so each request extends the previous one until the next compaction. A new compaction folds the previous summary into one capped summary (8,000 characters).
+- **Prefix contract**: Message-level prefix stability holds while the model, tools and `max_history = 0` stay unchanged and the history is append-only. Rewriting history (`/clear`, a compact, loading a session) starts a new generation: the summary is dropped and compaction waits for a fresh token count. This guarantees a stable message prefix; whether the server reuses its cache is up to the server.
 - **Native SDK**: Uses `ollama.AsyncClient` directly (not the OpenAI compatibility layer) for full timing data, native thinking support, and `keep_alive=-1` to prevent model unloading.
 
 **General-purpose manager** (Anthropic, OpenAI, Gemini, and other drivers):
 
-Every message is assigned an importance tier (system > recent user > recent assistant > tool results > old history). When the conversation approaches the token budget, old tool results are compressed (JSON structurally summarized, large text head/tail truncated) and the lowest-priority messages are pruned. Tool call + result pairs are kept as atomic units.
+Every message is assigned an importance tier (system > recent user > recent assistant > tool results > old history). When the conversation approaches the token budget, old tool results are compressed (JSON structurally summarized, large text head/tail truncated) and the lowest-priority messages are pruned. Tool call + result pairs are kept as atomic units. The newest unit is always kept, even when it alone exceeds the budget, so the model always sees the request it is answering.
 
 The context manager is configured under `[context_manager]` in `config.toml`:
 
@@ -451,7 +453,7 @@ You can also manually manage context with the `context` tool and its slash-comma
 | `/tasks` | Browse and implement tasks from `.ayder/tasks/` |
 | `/implement [id]` | Interactive task picker, or implement by ID (e.g., `/implement 1`) |
 | `/notes` | Browse and edit markdown notes |
-| `/skill` | Activate a domain skill from `.ayder/skills/` via the shared `skill` tool backend |
+| `/skill` | Load a skill as a user message: `/skill <name> [request]`, or pick one |
 | `/verbose` | Toggle verbose mode |
 | `/logging` | Set log level for current session (NONE, ERROR, WARNING, INFO, DEBUG, TRACE) |
 | `/compact` | Summarize conversation, save a context snapshot, clear, and reload |
@@ -465,10 +467,47 @@ You can also manually manage context with the `context` tool and its slash-comma
 | `/agent <name> <task>` | Dispatch an agent to run a task in the background |
 | `/agent cancel <name>` | Cancel a running agent |
 
-### Logging
+### Skills
+
+A skill is a directory with a `SKILL.md` file, optionally starting with
+`name:` and `description:` frontmatter. ayder looks in three places, and a
+later one overrides an earlier one with the same name:
+
+1. built-in skills in `~/.ayder/builtin/skills/`, managed by ayder and
+   replaced on upgrade; to customize one, copy it to `~/.ayder/skills/<name>/`
+2. your skills in `~/.ayder/skills/<name>/`
+3. project skills in `<project>/.ayder/skills/<name>/`
+
+ayder ships an `engineering` skill (build, change, debug or review code;
+research; plan multi-step work). The system prompt is a general assistant;
+when a skill fits the request, the model loads it itself.
+
+The skills enabled for a session are listed in the system prompt under
+"### Skills" (name and description), and only those can be loaded by the
+model. The `skills` setting chooses them:
+
+| `skills` | Enabled skills |
+|---|---|
+| not set | the built-in skills (`engineering`), including your or the project's copy of the same name |
+| `"none"` or `[]` | none: no skill tool, no catalog |
+| `["a", "b"]` | exactly these names, from any layer |
+
+The main session reads `[llm.<provider>].skills`, then `[app].skills`. An
+agent reads `[agents.<name>].skills`, then its provider profile's; never
+`[app]`. The `/skill` command loads any discovered skill, including project
+skills that are not in the catalog.
+
+The model loads one with `skill(action="load", name=...)`; the instructions
+come back as the tool result, so the conversation's earlier messages (and the
+provider's KV cache) are unchanged. Several skills can be active; other files
+in a skill's directory are read on demand with `skill(action="read")`.
+A skill directory or `SKILL.md` that links outside its skills directory is
+listed but never read or loaded; copy the skill instead of symlinking it.
+A loaded skill stays in context for the whole session: when trimming or
+compaction removes it, ayder adds it back right after the system prompt, or
+inside the Ollama summary.
 
-The LLM can also call `skill(action="list" | "load" | "unload")`
-automatically when a request clearly matches an available project skill.
+### Logging
 
 - Errors are always captured to `.ayder/log/errors.log`, even when
   `logging_level` is `NONE`. Set `file_enabled = false` under `[logging]` to
@@ -545,6 +584,8 @@ ayder-cli includes a built-in task system:
 
 Tasks are stored as markdown files in `.ayder/tasks/` using slug-based filenames (e.g., `TASK-001-add-auth-middleware.md`).
 
+The `task` tool resolves identifiers (a path, filename, ID or slug) only to files inside `.ayder/tasks/`. An identifier that points elsewhere, including through a symlink, is not resolved. If `.ayder/tasks/`, `.ayder/context/` or `.ayder/notes/` itself resolves outside the project, those tools return a security error.
+
 ```
 > /tasks            # Interactive task browser
 > /task-edit 1      # Open TASK-001 in the in-app editor
@@ -667,6 +708,7 @@ Each agent field:
 | `system_prompt` | Yes | The agent's role and instructions |
 | `provider` | No | LLM provider profile name (inherits from `[app]` if omitted) |
 | `model` | No | Model name override (inherits from provider profile if omitted) |
+| `skills` | No | `"none"` or a list of skill names (inherits from the provider profile; default: the built-in skills) |
 
 ### Using Agents
 

{ayder_cli-2.4.0 → ayder_cli-2.5.0}/README.md

@@ -161,10 +161,11 @@ editor = "vim"                # Editor for /task-edit command
 verbose = false               # Show file contents after write + LLM debug
 max_background_processes = 5  # Max concurrent background processes (1-20)
 max_iterations = 10           # Max agentic iterations per message (1-100)
-max_output_tokens = 4096      # Max tokens in LLM response
+max_output_tokens = 32768     # Max tokens in LLM response
 max_history_messages = 30     # Messages kept in history
 prompt = "STANDARD"           # System prompt tier: MINIMAL, STANDARD, EXTENDED
 tool_tags = ["core", "metadata"]  # Enabled tool tags (see /plugin)
+# skills = ["engineering"]    # or "none"; default: the built-in skills
 agent_timeout = 300           # Seconds before a background agent is cancelled
 
 [logging]
@@ -310,7 +311,7 @@ ayder -f instructions.txt
 ayder --stdin < prompt.txt
 
 # Override the main LLM system prompt with a file
-# (tool instructions + project structure are still appended)
+# (tool instructions are still appended)
 ayder --system-prompt prompt-file.md "refactor this code"
 ```
 
@@ -371,12 +372,13 @@ ayder-cli includes a context manager that automatically keeps your conversation
 - **Real token counts**: Uses Ollama's `prompt_eval_count` and `eval_count` from native API responses instead of estimates. Compaction decisions are based on actual usage, not heuristics.
 - **Cache monitoring**: Tracks `prompt_eval_duration / prompt_eval_count` ratio to detect KV-cache hits (hot/warm/miss). When the cache is hot, compaction is delayed (threshold raised from 70% to 90%) to avoid invalidating it.
 - **Auto-detected context length**: On startup, queries `/api/show` to get the model's real context length instead of relying on `num_ctx`. Falls back to `num_ctx` if detection fails.
-- **Heuristic compaction**: When the context fills up, oldest message units are replaced with a text summary — no LLM call needed.
+- **Heuristic compaction**: When the context fills up, the oldest half of the message units is replaced with a text summary — no LLM call needed. The newest unit is never folded. The summary persists: every later request reuses the same summary message, so each request extends the previous one until the next compaction. A new compaction folds the previous summary into one capped summary (8,000 characters).
+- **Prefix contract**: Message-level prefix stability holds while the model, tools and `max_history = 0` stay unchanged and the history is append-only. Rewriting history (`/clear`, a compact, loading a session) starts a new generation: the summary is dropped and compaction waits for a fresh token count. This guarantees a stable message prefix; whether the server reuses its cache is up to the server.
 - **Native SDK**: Uses `ollama.AsyncClient` directly (not the OpenAI compatibility layer) for full timing data, native thinking support, and `keep_alive=-1` to prevent model unloading.
 
 **General-purpose manager** (Anthropic, OpenAI, Gemini, and other drivers):
 
-Every message is assigned an importance tier (system > recent user > recent assistant > tool results > old history). When the conversation approaches the token budget, old tool results are compressed (JSON structurally summarized, large text head/tail truncated) and the lowest-priority messages are pruned. Tool call + result pairs are kept as atomic units.
+Every message is assigned an importance tier (system > recent user > recent assistant > tool results > old history). When the conversation approaches the token budget, old tool results are compressed (JSON structurally summarized, large text head/tail truncated) and the lowest-priority messages are pruned. Tool call + result pairs are kept as atomic units. The newest unit is always kept, even when it alone exceeds the budget, so the model always sees the request it is answering.
 
 The context manager is configured under `[context_manager]` in `config.toml`:
 
@@ -408,7 +410,7 @@ You can also manually manage context with the `context` tool and its slash-comma
 | `/tasks` | Browse and implement tasks from `.ayder/tasks/` |
 | `/implement [id]` | Interactive task picker, or implement by ID (e.g., `/implement 1`) |
 | `/notes` | Browse and edit markdown notes |
-| `/skill` | Activate a domain skill from `.ayder/skills/` via the shared `skill` tool backend |
+| `/skill` | Load a skill as a user message: `/skill <name> [request]`, or pick one |
 | `/verbose` | Toggle verbose mode |
 | `/logging` | Set log level for current session (NONE, ERROR, WARNING, INFO, DEBUG, TRACE) |
 | `/compact` | Summarize conversation, save a context snapshot, clear, and reload |
@@ -422,10 +424,47 @@ You can also manually manage context with the `context` tool and its slash-comma
 | `/agent <name> <task>` | Dispatch an agent to run a task in the background |
 | `/agent cancel <name>` | Cancel a running agent |
 
-### Logging
+### Skills
+
+A skill is a directory with a `SKILL.md` file, optionally starting with
+`name:` and `description:` frontmatter. ayder looks in three places, and a
+later one overrides an earlier one with the same name:
+
+1. built-in skills in `~/.ayder/builtin/skills/`, managed by ayder and
+   replaced on upgrade; to customize one, copy it to `~/.ayder/skills/<name>/`
+2. your skills in `~/.ayder/skills/<name>/`
+3. project skills in `<project>/.ayder/skills/<name>/`
+
+ayder ships an `engineering` skill (build, change, debug or review code;
+research; plan multi-step work). The system prompt is a general assistant;
+when a skill fits the request, the model loads it itself.
+
+The skills enabled for a session are listed in the system prompt under
+"### Skills" (name and description), and only those can be loaded by the
+model. The `skills` setting chooses them:
+
+| `skills` | Enabled skills |
+|---|---|
+| not set | the built-in skills (`engineering`), including your or the project's copy of the same name |
+| `"none"` or `[]` | none: no skill tool, no catalog |
+| `["a", "b"]` | exactly these names, from any layer |
+
+The main session reads `[llm.<provider>].skills`, then `[app].skills`. An
+agent reads `[agents.<name>].skills`, then its provider profile's; never
+`[app]`. The `/skill` command loads any discovered skill, including project
+skills that are not in the catalog.
+
+The model loads one with `skill(action="load", name=...)`; the instructions
+come back as the tool result, so the conversation's earlier messages (and the
+provider's KV cache) are unchanged. Several skills can be active; other files
+in a skill's directory are read on demand with `skill(action="read")`.
+A skill directory or `SKILL.md` that links outside its skills directory is
+listed but never read or loaded; copy the skill instead of symlinking it.
+A loaded skill stays in context for the whole session: when trimming or
+compaction removes it, ayder adds it back right after the system prompt, or
+inside the Ollama summary.
 
-The LLM can also call `skill(action="list" | "load" | "unload")`
-automatically when a request clearly matches an available project skill.
+### Logging
 
 - Errors are always captured to `.ayder/log/errors.log`, even when
   `logging_level` is `NONE`. Set `file_enabled = false` under `[logging]` to
@@ -502,6 +541,8 @@ ayder-cli includes a built-in task system:
 
 Tasks are stored as markdown files in `.ayder/tasks/` using slug-based filenames (e.g., `TASK-001-add-auth-middleware.md`).
 
+The `task` tool resolves identifiers (a path, filename, ID or slug) only to files inside `.ayder/tasks/`. An identifier that points elsewhere, including through a symlink, is not resolved. If `.ayder/tasks/`, `.ayder/context/` or `.ayder/notes/` itself resolves outside the project, those tools return a security error.
+
 ```
 > /tasks            # Interactive task browser
 > /task-edit 1      # Open TASK-001 in the in-app editor
@@ -624,6 +665,7 @@ Each agent field:
 | `system_prompt` | Yes | The agent's role and instructions |
 | `provider` | No | LLM provider profile name (inherits from `[app]` if omitted) |
 | `model` | No | Model name override (inherits from provider profile if omitted) |
+| `skills` | No | `"none"` or a list of skill names (inherits from the provider profile; default: the built-in skills) |
 
 ### Using Agents
 

{ayder_cli-2.4.0 → ayder_cli-2.5.0}/docs/GATE.md

@@ -12,7 +12,7 @@ Input: the candidate SHA.
    gates and the test suite, and stops at the first failing task.
 4. Copy the final pytest summary line (`=== N passed, M skipped in …s ===`) verbatim.
 
-Last measured: `2494 passed, 2 skipped` on `feat/evren-terms` with the evren default profile (2026-09-29).
+Last measured: `2707 passed, 2 skipped` on `chore/prompt-tests-injection-only` (2026-10-16).
 
 ## Report format
 

{ayder_cli-2.4.0 → ayder_cli-2.5.0}/docs/PROJECT_STRUCTURE.md

@@ -72,6 +72,7 @@ ayder-cli is an AI agent chat client with a modular, layered architecture:
 │  core/config.py              (Pydantic Config)                 │
 │  core/context.py             (ProjectContext, sandboxing)      │
 │  core/result.py              (ToolSuccess, ToolError)          │
+│  core/skills.py              (skill discovery, delivery)       │
 │  core/context_manager.py     (ContextManagerProtocol)          │
 │  core/default_context_manager.py                               │
 │  core/ollama_context_manager.py  (KV-cache-aware)              │
@@ -154,6 +155,8 @@ cli.py:main()
 | `core/config_migration.py` | TOML config migration helpers | migration routines for schema changes |
 | `core/context.py` | Project sandboxing | `ProjectContext` |
 | `core/result.py` | Tool result types | `ToolSuccess`, `ToolError` |
+| `core/skills.py` | Skill discovery, delivery, built-in sync and the session filter | `SkillInfo`, `SkillScope`, `discover_all()`, `render_skill_block()`, `active_skills()`, `builtin_skills_root()`, `build_scope()`, `format_catalog()` |
+| `skills/<name>/SKILL.md` | Packaged built-in skills (data only, shipped in the wheel); synced to `~/.ayder/builtin/skills` when `~/.ayder/builtin/STAMP` (version + SHA-256) differs | - |
 | `core/context_manager.py` | ContextManager Protocol + shared utilities | `ContextManagerProtocol`, `ContextStats`, `truncate_tool_result()` |
 | `core/default_context_manager.py` | Tiered context manager for non-Ollama providers | `DefaultContextManager`, `TokenCounter`, `MessageTier` |
 | `core/ollama_context_manager.py` | KV-cache-aware context manager for Ollama | `OllamaContextManager`, `OllamaContextStats` |
@@ -294,6 +297,7 @@ the design rationale.
 | `tools/builtins/context.py` | Session context snapshots | `context` |
 | `tools/builtins/notes.py` | Note-taking tool | `create_note()` |
 | `tools/builtins/tasks.py` | Task-management tools | `list_tasks()`, `show_task()` |
+| `tools/builtins/skill.py` | Skill tool impl | `skill()`, `discover_skills()` |
 | `tools/builtins/web.py` | HTTP fetch tool | `fetch_web()` |
 | `tools/builtins/utils_tools.py` | Misc utility tools | `manage_environment_vars()` |
 | `tools/builtins/*_definitions.py` | Per-domain tool definitions (9 files) | `TOOL_DEFINITIONS` tuples auto-discovered at import |
@@ -303,7 +307,7 @@ the design rationale.
 | Module | Purpose | Key Classes/Functions |
 |--------|---------|----------------------|
 | `process_manager.py` | Background processes | `ProcessManager`, `background_process()` |
-| `prompts.py` | System prompts | `SYSTEM_PROMPT`, `PLANNING_PROMPT_TEMPLATE`, `get_system_prompt()`, `PROJECT_STRUCTURE_MACRO_TEMPLATE` |
+| `prompts.py` | System prompts | `SYSTEM_PROMPT`, `PLANNING_PROMPT_TEMPLATE`, `get_system_prompt()` |
 | `parser.py` | XML tool-call parsing | `parse_custom_tool_calls()` |
 
 ### Theme Modules
@@ -590,7 +594,7 @@ AyderApp is the main Textual application implementing a chat-style interface:
 The application manages:
 - LLM provider setup and configuration
 - Tool registry initialization with middleware
-- System prompt construction with project structure
+- System prompt construction (base prompt + tool prompts + skill catalog; no project tree)
 - Process manager for background operations
 
 ### Chat Loop Summary (src/ayder_cli/loops/chat_loop.py)
@@ -605,9 +609,11 @@ previously-split `AgentLoopBase` and `LoopConfig` were merged back in):
 - Routes *every* LLM response through the injected `ContextManagerProtocol` (`prepare_messages()` / `update_from_response()`); tool outputs are truncated at insertion time via `truncate_tool_result()` so the KV-cache prefix remains stable
 
 Key features:
-- Auto-approved tools run in parallel (`asyncio.gather`); confirmation-required tools run sequentially
+- Auto-approved tools run in parallel (`asyncio.as_completed`); confirmation-required tools run sequentially
 - Confirmation gate applies to **both** OpenAI-format and XML/JSON custom tool calls
-- `ExecutionPolicy.execute_with_registry()` is the single execution entry (validate → permission → execute)
+- `ExecutionPolicy.execute_with_registry()` is the single execution entry (object check → `canonicalize_arguments()` → validate → permission → execute). Canonicalization applies aliases and `"123"`/`"true"`/`"false"` coercion; the loop's required-argument gate uses the same function
+- Argument integrity: `_parse_tool_arguments()` reports `ok`, `repaired` or `invalid`. Non-object arguments never run. Arguments repaired after being cut off (output token limit) run only for tools registered explicitly as `r`; other tools get an error asking the model to resend in smaller calls
+- Cancellation: each batch owns a `_ToolBatch`. Workers refuse to start once the batch is cancelled and record their own result and `tool_call` event. When a batch ends early (Ctrl+C or a raising callback), a `finally` appends exactly one tool message per unanswered call: its result, an "interrupted, may have partially applied" notice, or a "cancelled before it ran" notice. Running workers are not awaited, so Ctrl+C returns at once
 - Token usage tracking and iteration limiting
 - Pre-iteration hook (`pre_iteration_hook`) used by the agent system to inject summaries
 - Graceful cancellation via `is_cancelled()`
@@ -717,6 +723,8 @@ Schema-driven tool definitions with auto-discovery:
 - `OllamaContextManager` (`core/ollama_context_manager.py`) — stable-prefix + real-token accounting, uses `prompt_eval_count` / `prompt_eval_duration` directly.
 - `CacheMonitor` (`core/cache_monitor.py`) — timing-based KV-cache hit detection driving adaptive compaction thresholds.
 
+Both managers re-supply active skills that a trim or compaction removed (`core/skills.py` `active_skills`).
+
 Selection lives in `core/context_manager_factory.py` (driver → class path registry). `create_runtime()` owns instantiation and calls `freeze_system_prompt()` once.
 
 ### Phase — opus47 critical fixes (2026-04-16)

{ayder_cli-2.4.0 → ayder_cli-2.5.0}/docs/config.toml.example

@@ -97,6 +97,10 @@ agent_timeout = 1800               # seconds before a background agent is cancel
 # tasks/notes (metadata), background process control for test suites
 # (background), and web fetch (http). See /plugin for the full tag list.
 tool_tags = ["core", "metadata", "background", "http"]
+# Skills the session's model may load (listed under "### Skills" in the system
+# prompt). Not set = the built-in skills; "none" = no skill tool; or a strict
+# list. A [llm.*] profile's skills wins over this one for the session.
+# skills = ["engineering"]
 
 [ui]
 # TUI layout (theme) — picks the chrome/layout:
@@ -180,6 +184,7 @@ driver = "ollama"
 base_url = "http://localhost:11434"
 model = "qwen3.6:35b-a3b-coding-nvfp4"
 num_ctx = 32768                     # keep modest: local VRAM is the constraint
+# skills = "none"                   # wins over [app].skills; agents on this profile inherit it
 reasoning_effort = "default"
 think = true                        # default; set false if the selected model rejects thinking
 
@@ -201,7 +206,7 @@ think = true                        # default; set false if the selected model r
 # api_key = "sk-..."                        # your key (or "ollama"/"none" for local servers)
 # model = "gpt-5.5"
 # num_ctx = 65536            # sizes the context-manager budget (default 65536); NOT sent to the API
-# max_output_tokens = 8192   # -> max_tokens / max_completion_tokens (default 4096); keep WELL below num_ctx
+# max_output_tokens = 8192   # -> max_tokens / max_completion_tokens (default 32768); keep WELL below num_ctx
 # stop_sequences = []        # -> stop (default [])
 # reasoning_effort = "medium" # choose a value supported by the selected model
 # # temperature / think are NOT applied by the openai driver.
@@ -224,6 +229,7 @@ think = true                        # default; set false if the selected model r
 [agents.pm_spec]
 provider = "ollama_cloud"
 model = "glm-5.2:cloud"
+# skills = ["engineering"]          # wins over the provider profile's skills; [app] is never read
 temperature = 0.4
 max_output_tokens = 8192
 reasoning_effort = "high"

{ayder_cli-2.4.0 → ayder_cli-2.5.0}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "ayder-cli"
-version = "2.4.0"
+version = "2.5.0"
 description = "AI agent for any LLMs"
 readme = "README.md"
 requires-python = ">=3.12"

{ayder_cli-2.4.0 → ayder_cli-2.5.0}/scripts/check_exception_handlers.py

@@ -24,7 +24,7 @@ DEFAULT_BASELINE = REPO / "scripts" / "exception_baseline.txt"
 
 # Frozen census (§C10). A different number means the gate is looking at the
 # wrong tree, or the census moved without the plan being updated.
-EXPECTED_BROAD = 137
+EXPECTED_BROAD = 134
 BRANCHING = (ast.If, ast.Try, ast.For, ast.While, ast.With, ast.Match)
 TERMINAL = (ast.Return, ast.Break, ast.Continue, ast.Pass)
 

{ayder_cli-2.4.0 → ayder_cli-2.5.0}/scripts/check_log_arguments.py

@@ -203,13 +203,13 @@ FROZEN_CHAIN_SITES: dict[tuple[str, str, str, str], tuple[str, int]] = {
     ("log.py", "emit_event", "bind", "**"): ("Name(id='fields', ctx=Load())", 1),
 }
 
-# The 101 `exception=True` inputs are homogeneous, so they are governed as one
-# package-wide shape/count rule rather than 101 site identities - their per-site
+# The 98 `exception=True` inputs are homogeneous, so they are governed as one
+# package-wide shape/count rule rather than 98 site identities - their per-site
 # discipline is C10's.
-FROZEN_EXCEPTION_TRUE = 101
+FROZEN_EXCEPTION_TRUE = 98
 
 # Live chain inputs, counted per method. Printed on every run.
-CHAIN_TALLY: dict[str, int] = {"bind": 6, "opt": 105, "patch": 0}
+CHAIN_TALLY: dict[str, int] = {"bind": 6, "opt": 102, "patch": 0}
 
 # The reviewed composition of THE COMMITTED baseline, frozen by the Phase 5
 # census: 252 retained non-path + 22 paths + 2 deferred interpolated rows, plus
@@ -218,12 +218,12 @@ CHAIN_TALLY: dict[str, int] = {"bind": 6, "opt": 105, "patch": 0}
 # unchanged - so the composition is gated too. Caller-supplied baselines (the
 # gate's own synthetic fixtures) are exempt: this is a fact about one file.
 FROZEN_TALLY: dict[str, int] = {
-    "R-name": 96,
+    "R-name": 99,
     "R-id": 40,
     "R-count": 90,
     "R-status": 27,
-    "R-class": 10,
-    "R-path": 24,
+    "R-class": 11,
+    "R-path": 25,
     # Both 5-04 deferrals are closed: the two asyncio rows now interpolate the
     # B2-prime normalized value, and the bridge's dynamic message is masked by
     # the prose sinks it reaches.

{ayder_cli-2.4.0 → ayder_cli-2.5.0}/scripts/log_argument_baseline.txt

@@ -83,6 +83,7 @@
 {"class":"R-name","site":{"arg_index":"pos:0","expr":"name","message":"Uninstalled plugin '{}'","method":"info","path":"tools/plugin_manager.py","qualname":"uninstall_plugin"}}
 {"class":"R-name","site":{"arg_index":"pos:0","expr":"name","message":"cancel: agent='{}' \u2014 no running/queued instances","method":"debug","path":"agents/registry.py","qualname":"AgentRegistry.cancel"}}
 {"class":"R-name","site":{"arg_index":"pos:0","expr":"name","message":"cancel: agent='{}' cancelled {} instance(s): {}","method":"debug","path":"agents/registry.py","qualname":"AgentRegistry.cancel"}}
+{"class":"R-name","site":{"arg_index":"pos:0","expr":"name","message":"skills lists {!r}, but no skill directory provides it","method":"warning","path":"core/skills.py","qualname":"build_scope"}}
 {"class":"R-status","site":{"arg_index":"pos:0","expr":"normalized","message":"Unhandled asyncio error: {}","method":"error","path":"diagnostics.py","qualname":"_handle_asyncio"}}
 {"class":"R-status","site":{"arg_index":"pos:0","expr":"normalized","message":"Unhandled asyncio exception: {}","method":"critical","path":"diagnostics.py","qualname":"_handle_asyncio"}}
 {"class":"R-count","site":{"arg_index":"pos:0","expr":"num_ctx","message":"Context budget: {} tokens (derived from num_ctx; max_context_tokens unset)","method":"info","path":"core/default_context_manager.py","qualname":"DefaultContextManager.__init__"}}
@@ -138,7 +139,9 @@
 {"class":"R-status","site":{"arg_index":"pos:0","expr":"self.config.chat_protocol","message":"Ollama chat_protocol={!r} is not recognized; forcing generic_xml fallback","method":"warning","path":"providers/impl/ollama.py","qualname":"OllamaProvider.stream_with_tools"}}
 {"class":"R-name","site":{"arg_index":"pos:0","expr":"self.config.model","message":"LLM returned empty response (possible connection drop). model={}, provider={}","method":"error","path":"loops/chat_loop.py","qualname":"ChatLoop.run"}}
 {"class":"R-status","site":{"arg_index":"pos:0","expr":"signal.Signals(signum).name","message":"Received signal {}; flushing logs","method":"warning","path":"diagnostics.py","qualname":"install_exception_hooks._chained_signal"}}
+{"class":"R-name","site":{"arg_index":"pos:0","expr":"skill_dir.name","message":"Skill directory {!r} declares name {!r}; the directory name is used","method":"debug","path":"core/skills.py","qualname":"read_skill_info"}}
 {"class":"R-name","site":{"arg_index":"pos:0","expr":"source.repo","message":"Downloading '{}' at {}","method":"info","path":"tools/plugin_github.py","qualname":"download_plugin"}}
+{"class":"R-path","site":{"arg_index":"pos:0","expr":"target","message":"Built-in skills were not synced to {} ({}); reading the packaged copy","method":"warning","path":"core/skills.py","qualname":"_sync_builtin_skills"}}
 {"class":"R-name","site":{"arg_index":"pos:0","expr":"tc.function.name","message":"Tool '{}' called with missing args {} ({} chars raw)","method":"warning","path":"loops/chat_loop.py","qualname":"ChatLoop._execute_tool_calls"}}
 {"class":"R-name","site":{"arg_index":"pos:0","expr":"tc_entry['function'].get('name', '?')","message":"Malformed tool arguments for '{}': {} chars \u2014 repairing before storing in history","method":"warning","path":"loops/chat_loop.py","qualname":"ChatLoop.run"}}
 {"class":"R-name","site":{"arg_index":"pos:0","expr":"tc_obj.function.name","message":"Tool execution cancelled: {}","method":"warning","path":"loops/chat_loop.py","qualname":"ChatLoop._execute_tool_calls._safe_exec"}}
@@ -160,6 +163,7 @@
 {"class":"R-status","site":{"arg_index":"pos:1","expr":"', '.join(valid)","message":"Unsupported driver (value_chars={}); must be one of: {}. Fix the driver in your [llm.<provider>] profile.","method":"error","path":"core/config.py","qualname":"Config.validate_driver"}}
 {"class":"R-name","site":{"arg_index":"pos:1","expr":"agent_name","message":"Agent panel could not show run {} of agent {}","method":"debug","path":"tui/commands.py","qualname":"handle_agent"}}
 {"class":"R-id","site":{"arg_index":"pos:1","expr":"commit_sha[:7]","message":"Downloading '{}' at {}","method":"info","path":"tools/plugin_github.py","qualname":"download_plugin"}}
+{"class":"R-name","site":{"arg_index":"pos:1","expr":"declared","message":"Skill directory {!r} declares name {!r}; the directory name is used","method":"debug","path":"core/skills.py","qualname":"read_skill_info"}}
 {"class":"R-count","site":{"arg_index":"pos:1","expr":"delay","message":"Provider stream failed ({}); retrying in {:.2f}s ({} attempts left)","method":"info","path":"providers/retry.py","qualname":"RetryingProvider.stream_with_tools"}}
 {"class":"R-status","site":{"arg_index":"pos:1","expr":"driver.mode.value","message":"Ollama driver={} mode={} for {!r}","method":"info","path":"providers/impl/ollama.py","qualname":"OllamaProvider.stream_with_tools"}}
 {"class":"R-status","site":{"arg_index":"pos:1","expr":"driver.mode.value","message":"{} ({}) failed mid-stream (error_type={}); transparently retrying with {} ({})","method":"info","path":"providers/impl/ollama.py","qualname":"OllamaProvider.stream_with_tools"}}
@@ -222,6 +226,7 @@
 {"class":"R-name","site":{"arg_index":"pos:1","expr":"tc.name","message":"Tool call id reused by a different tool ({} then {}); calls would merge","method":"warning","path":"loops/chat_loop.py","qualname":"ChatLoop.run"}}
 {"class":"R-status","site":{"arg_index":"pos:1","expr":"think","message":"Ollama model {!r} rejected think={!r}; retrying with think=False","method":"info","path":"providers/impl/ollama.py","qualname":"OllamaProvider._stream_in_content"}}
 {"class":"R-status","site":{"arg_index":"pos:1","expr":"think","message":"Ollama model {!r} rejected think={!r}; retrying with think=False","method":"info","path":"providers/impl/ollama.py","qualname":"OllamaProvider._stream_native"}}
+{"class":"R-class","site":{"arg_index":"pos:1","expr":"type(exc).__name__","message":"Built-in skills were not synced to {} ({}); reading the packaged copy","method":"warning","path":"core/skills.py","qualname":"_sync_builtin_skills"}}
 {"class":"R-class","site":{"arg_index":"pos:1","expr":"type(exc).__name__","message":"Provider retry budget exhausted after {} attempts; raising {}","method":"warning","path":"providers/retry.py","qualname":"RetryingProvider.stream_with_tools"}}
 {"class":"R-class","site":{"arg_index":"pos:1","expr":"type(exc).__name__","message":"Unreadable context slot at {}: {}","method":"warning","path":"tools/builtins/context.py","qualname":"_scan_context_dir"}}
 {"class":"R-name","site":{"arg_index":"pos:2","expr":"', '.join(discovered_modules)","message":"Auto-discovered {} tool definitions from {} modules: {}","method":"info","path":"tools/definition.py","qualname":"_discover_definitions"}}

{ayder_cli-2.4.0 → ayder_cli-2.5.0}/src/ayder_cli/agents/config.py

@@ -1,9 +1,10 @@
 """AgentConfig — Pydantic model for agent definitions."""
 
-from typing import Any
+from typing import Any, Literal
 
-from pydantic import BaseModel, ConfigDict, model_validator
+from pydantic import BaseModel, ConfigDict, field_validator, model_validator
 
+from ayder_cli.core.config import _coerce_skills_setting
 from ayder_cli.core.reasoning import ReasoningEffort, ThinkOption
 
 
@@ -32,6 +33,13 @@ class AgentConfig(BaseModel):
     max_output_tokens: int | None = None
     think: ThinkOption = None
     reasoning_effort: ReasoningEffort = None
+    # None: the provider profile's setting; "none" or a list: see Config.skills.
+    skills: list[str] | Literal["none"] | None = None
+
+    @field_validator("skills", mode="before")
+    @classmethod
+    def _validate_skills(cls, v: Any) -> Any:
+        return _coerce_skills_setting(v)
 
     @model_validator(mode="before")
     @classmethod

{ayder_cli-2.4.0 → ayder_cli-2.5.0}/src/ayder_cli/application/execution_policy.py

@@ -158,19 +158,40 @@ class ExecutionPolicy:
         *,
         pre_approved: bool = False,
     ) -> ExecutionResult:
-        """Single shared execution path: validate → check permission → execute.
+        """Single shared execution path: canonicalize → validate → permission → execute.
 
         Both CLI and TUI route through this method so registry.execute is never
         called directly from interface-specific code.
 
+        Order:
+        1. Arguments that are not a JSON object are a validation error.
+        2. ``canonicalize_arguments``: aliases, integer and boolean coercion.
+        3. ``ValidationAuthority`` on the canonical arguments.
+        4. Permission check (unless pre-approved), then ``registry.execute``
+           with the canonical arguments.
+
         Args:
             pre_approved: If True, skip the permission check (user already
                 confirmed via the confirmation dialog).
         """
         from ayder_cli.application.validation import ValidationAuthority
+        from ayder_cli.tools.normalization import canonicalize_arguments
+
+        if not isinstance(request.arguments, dict):
+            return ExecutionResult(
+                success=False,
+                error=ValidationError(
+                    tool_name=request.name,
+                    message=(
+                        "arguments must be a JSON object, got "
+                        f"{type(request.arguments).__name__}"
+                    ),
+                ),
+            )
+        canonical = canonicalize_arguments(request.name, request.arguments)
 
         authority = ValidationAuthority()
-        v_request = ToolRequest(name=request.name, arguments=request.arguments)
+        v_request = ToolRequest(name=request.name, arguments=canonical)
         valid, err = authority.validate(v_request)
         if not valid:
             return ExecutionResult(success=False, error=err)
@@ -180,7 +201,7 @@ class ExecutionPolicy:
             if perm_err is not None:
                 return ExecutionResult(success=False, error=perm_err)
 
-        raw = registry.execute(request.name, request.arguments)
+        raw = registry.execute(request.name, v_request.arguments)
         if isinstance(raw, ToolError):
             # A tool that ran and failed is NOT a successful execution.
             return ExecutionResult(success=False, was_confirmed=pre_approved, error=raw)

{ayder_cli-2.4.0 → ayder_cli-2.5.0}/src/ayder_cli/application/runtime_factory.py

@@ -7,6 +7,7 @@ eliminating the duplicate `_build_services()` / `AyderApp.__init__()` wiring.
 from __future__ import annotations
 
 from dataclasses import dataclass, field
+from pathlib import Path
 from typing import TYPE_CHECKING, Any
 
 from ayder_cli.core.config import Config, load_config, load_config_for_provider
@@ -16,13 +17,13 @@ from ayder_cli.core.context_manager_factory import context_manager_factory
 if TYPE_CHECKING:
     from ayder_cli.agents.config import AgentConfig
 from ayder_cli.core.context import ProjectContext
+from ayder_cli.core.skills import build_scope
 from ayder_cli.providers import AIProvider, provider_orchestrator
 from ayder_cli.providers.retry import RetryConfig, RetryingProvider
 from ayder_cli.tools.registry import ToolRegistry, create_default_registry
 from ayder_cli.process_manager import ProcessManager
 from ayder_cli.prompts import (
     get_system_prompt,
-    PROJECT_STRUCTURE_MACRO_TEMPLATE,
 )
 from ayder_cli.log import get_logger
 
@@ -42,6 +43,25 @@ class RuntimeComponents:
     context_manager: Any = field(default=None)  # ContextManagerProtocol
 
 
+def apply_skill_setting(
+    tool_registry: ToolRegistry,
+    project_ctx: ProjectContext,
+    setting: Any,
+    messages: list[dict] | None = None,
+) -> None:
+    """Scope the registry's skills; hide the skill tool for "none", show it otherwise.
+
+    Called at runtime construction and again on a provider switch, where
+    ``messages`` is the live history (ChatLoop binds it only once).
+    """
+    tool_registry.skill_scope = build_scope(Path(project_ctx.root), setting)
+    tool_registry.skill_scope.messages = messages
+    if setting == "none":
+        tool_registry.excluded_tools.add("skill")
+    else:
+        tool_registry.excluded_tools.discard("skill")
+
+
 def create_runtime(
     *,
     config: Config | None = None,
@@ -60,7 +80,7 @@ def create_runtime(
             (e.g. "AGENTIC" for `ayder-cli --agent`). See prompts.get_system_prompt.
         system_prompt_override: When set, use this text as the system-prompt base
             instead of prompts.get_system_prompt (ayder --system-prompt FILE). Tool
-            prompts and the project-structure macro are still appended.
+            prompts are still appended.
 
     Returns:
         RuntimeComponents with all dependencies wired.
@@ -84,15 +104,7 @@ def create_runtime(
         process_manager=process_manager,
         context_manager=context_mgr,
     )
-
-    try:
-        structure = tool_registry.execute("get_project_structure", {"max_depth": 3})
-        macro = PROJECT_STRUCTURE_MACRO_TEMPLATE.format(project_structure=structure)
-    except Exception:  # noqa: BLE001 - tool boundary: the project-structure macro is optional, so runtime construction continues without it
-        logger.opt(exception=True).warning(
-            "Project-structure macro unavailable; continuing without it"
-        )
-        macro = ""
+    apply_skill_setting(tool_registry, project_ctx, cfg.session_skills())
 
     # Format the final prompt (protocol injection now handled dynamically by chat_loop and protocols)
     if system_prompt_override is not None:
@@ -101,7 +113,7 @@ def create_runtime(
         base_prompt = get_system_prompt(cfg.prompt)
     tool_tags = frozenset(cfg.tool_tags) if cfg.tool_tags else None
     tool_prompts = tool_registry.get_system_prompts(tags=tool_tags)
-    system_prompt = base_prompt + tool_prompts + macro
+    system_prompt = base_prompt + tool_prompts
 
     # Freeze system prompt after tool schemas are known.
     tool_schemas = tool_registry.get_schemas(tags=tool_tags)
@@ -196,6 +208,9 @@ def create_agent_runtime(
         process_manager=process_manager,
         context_manager=context_mgr,
     )
+    # An agent reads its own setting, then its provider profile's; never [app].
+    skills_setting = agent_config.skills if agent_config.skills is not None else cfg.profile_skills
+    apply_skill_setting(tool_registry, project_ctx, skills_setting)
 
     # 5. Build agent system prompt
     identity_prefix = (