Using the tracedecay CLI

MCP is optional. The tracedecay binary exposes every MCP tool as a first-class shell command with the same project store, arguments, and payloads. Use the CLI directly when MCP is intentionally unavailable, or switch to it after a transport failure, and keep following whatever tracedecay:* skill you were in.

The one rule for arguments

The arguments of tracedecay tool <name> are the tool's MCP arguments object — the same JSON you would send over MCP. Pass it whole with --args:

• --args - + a quoted heredoc — the canonical form. No shell escaping, no argv-length limit, byte-exact MCP parity. Use it whenever the payload has quotes, newlines, or any non-scalar value.
• --args '<json>' — inline JSON, only when the object is short and contains no single quotes.
• --args payload.json (or --args @payload.json) — read from a file; use for payloads near or over the ~128 KiB per-argument shell limit.

tracedecay tool multi_str_replace --args - <<'JSON'
{"path":"src/lib.rs","replacements":[["old, with 'quotes'","new $body"]]}
JSON

Nothing new to learn: you already know every tool's MCP schema, and --args carries that object verbatim. For quick scalar-only calls you may also spell top-level fields as --key value flags (--key=value works too); values are interpreted by the tool's schema, and anything that isn't a scalar must be JSON. See references/tool-arg-catalog.md for ready-to-copy examples of the hard-shape tools.

Discovery

1. List every tool → tracedecay tool (no name): all tools grouped by category with one-line summaries.
2. One tool's parameters → tracedecay tool <name> --help: the tool's full description, a ready-to-copy usage line, each parameter with its type/required flag (enum values and array shapes included), and a generated --args - example for any tool whose payload is non-scalar.
3. Everything else → tracedecay --help: the non-tool subcommands (init, sync, status, doctor, daemon, sessions, dashboard, …) plus a quick-start trailer that restates this discovery flow. Every subcommand's own --help carries an Examples: section with real flag combinations and Related: cross-references — read it before improvising flags.

Pre-flighting edits with --dry-run

Add --dry-run to parse, validate, and print the final arguments object as JSON without dispatching the tool — no daemon, no handler, no side effects. Use it to self-check a destructive edit-tool payload (str_replace, multi_str_replace, replace_symbol, ast_grep_rewrite, insert_at) before applying it, or to confirm you constructed the right object when a shape is tricky. Exits non-zero with the same corrective error as a real call if the arguments are wrong.

tracedecay tool multi_str_replace --args - --dry-run <<'JSON'
{"path":"src/lib.rs","replacements":[["old","new"]]}
JSON

Reserved flags

• --args <json|-|@file|file> — whole MCP arguments object (see above). Mutually exclusive with --key value flags and positionals.
• --dry-run — validate + print the resolved arguments object, do not invoke the tool.
• --json — print the raw JSON result payload instead of the human-readable text rendering.
• --project <path> — pick the project root explicitly; otherwise the nearest initialised project walking up from cwd is used. (We use --project, not -p, because several tools have a path argument that filters files within the project.)
• -h / --help — print the tool's parameters and exit.
• Per-key values starting with @ are read from that file (--new-body @/tmp/body.txt); @- reads stdin — handy for multi-line replacement bodies.

Retrieving truncated responses

TraceDecay truncates large tool responses and emits a handle envelope instead of the full body. The original text is cached in the active-project store; dereference it rather than re-running the source tool — this works identically over MCP (tracedecay_retrieve) and the CLI (tracedecay tool retrieve).

• A prior response ended with a handle (e.g. rh_…) and the missing detail is actually needed → dereference the handle → tracedecay_retrieve (handle copied exactly). It returns the exact cached original text; it does not re-run the tool or re-read a file/session/node. Do not re-run the broad query, guess, or read a file again.
• You do NOT need the truncated tail → leave it; retrieval costs tokens.
• Handles are local, project-scoped, and expire; if retrieve reports an expired/unknown handle, re-run the original tool with a narrower query (see tracedecay:using-tracedecay) rather than retrying the stale handle. If the truncated response used a project-id/project-path selector, pass the same selector to retrieve.
• To open one session/summary node instead of a cached tool body → expand it with tracedecay_lcm_expand; tracedecay:managing-session-context drives the LCM store and past-session retrieval.

CLI fallback conditions

• MCP is intentionally unavailable or omitted from the host integration.
• An MCP call returns a client or transport error, times out, or the server drops mid-session.
• The tracedecay MCP server is not configured in this host but tracedecay is on PATH.
• A subagent or hook context has shell access but no MCP access.

For a transport failure, diagnose the MCP side with tracedecay doctor and tracedecay tool runtime. When CLI-only operation is intentional, do not treat the missing MCP transport as an error.

Corrective errors are the feedback loop

If a call is rejected, the error message tells you exactly how to fix it — read it as the CLI's tab-completion:

• Unknown parameter → names the nearest valid parameter (did you mean --limit?) and lists all valid keys.
• Invalid enum → lists the allowed values verbatim (not one of: complexity | lines | …).
• Wrong type for a field → names the expected JSON type and shows the --args - heredoc form for non-scalars.
• Missing required → names the parameter and gives a one-line usage example.

Fix the call per the message and retry; do not abandon the CLI fallback for raw DB access or broad Grep.

Guardrails

• Never query .tracedecay/*.db with sqlite3 or scripts — schemas are internal and change without notice. The CLI is the supported fallback, not raw DB access.
• Do not abandon tracedecay for broad Grep/file reads just because MCP transport failed; the CLI answers the same graph, memory, and session questions.
• CLI editing tools (str_replace, replace_symbol, …) mutate the working tree exactly like their MCP twins — apply the same care as tracedecay:editing-safely. Pre-flight with --dry-run when the payload is non-trivial.
• If the CLI also fails (binary missing or project not initialised), fall back to plain tools and suggest tracedecay init / tracedecay doctor to the user.

If tools are deferred or MCP fails

• This skill is the MCP-failure path: run tracedecay tool <name> (see The one rule for arguments above) for any tool whose MCP call errored, timed out, or was never configured.
• Deferred (names listed without schemas) but MCP otherwise healthy: load once with ToolSearch — select:tracedecay_retrieve,tracedecay_runtime (add the tools the parent skill needs) — then call normally instead of shelling out.

Deliverable

Do not end this workflow without: the same result the MCP tool would have returned, plus a note that the CLI fallback was used and why. Report any tracedecay_metrics: line to the user.